A System Dynamics Model of the Development of New ... · Appendix C. Evolution of Causal Loop...
Transcript of A System Dynamics Model of the Development of New ... · Appendix C. Evolution of Causal Loop...
A System Dynamics Model of the Development
of New Technologies for Ship Systems
Pavinder Monga
Thesis submitted to the faculty of the Virginia Polytechnic Institute and
State University in partial fulfillment of the requirements for the degree of
Master of Science
in
Industrial and Systems Engineering
K. P. Triantis, Ph.D., Chairman
W. G. Sullivan, Ph.D.
C. P. Koelling, Ph.D.
September 18, 2001
Falls Church, VA
Keywords: System Dynamics modeling, technology development, cost
overruns, dynamic behavior
Copyright 2001, Pavinder Monga
A System Dynamics Model of the Development
of New Technologies for Ship Systems
Pavinder Monga
(Abstract)
Key words: System Dynamics modeling, technology development, cost overruns,
dynamic behavior
System Dynamics has been applied to various fields in the natural and social sciences.
There still remain countless problems and issues where understanding is lacking and the
dominant theories are event-oriented rather than dynamic in nature. One such research
area is the application of the traditional systems engineering process in new technology
development. The Navy has been experiencing large cost overruns in projects dealing
with the implementation of new technologies on complex ship systems. We believe that
there is a lack of understanding of the dynamic nature of the technology development
process undertaken by aircraft-carrier builders and planners. Our research effort is to
better understand the dynamics prevalent in the new technology development process and
we use a dynamic modeling technique, namely, System Dynamics in our study.
iii
We provide a comprehensive knowledge elicitation process in which members from the
Newport News Shipbuilding, the Naval Sea Command Cost Estimating Group, and the
Virginia Tech System Performance Laboratory take part in a group model building
exercise. We build a System Dynamics model based on the information and data
obtained from the experts. Our investigation of the dynamics yields two dominant
behaviors that characterize the technology development process. These two dynamic
behaviors are damped oscillation and goal seeking. Furthermore, we propose and
investigate four dynamic hypotheses in the system. For the current structure of the
model, we see that an increase in the complexity of new technologies leads to an increase
in the total costs, whereas a increase in the technology maturity leads to a decrease in the
total costs in the technology development process. Another interesting insight is that an
increase in training leads to a decrease in total costs.
iv
Acknowledgements
I think research is a long and enriching walk. The path is sometimes hidden and very
difficult to find; at times it is right ahead, well formed and well illuminated. During my
research walk, I met a number of intellectuals. They contributed immensely in my
research efforts, providing me with most valuable ideas, and helping me find the path
every time I used to get lost during my walk.
I would like to express my utmost gratitude for my advisor, Dr. Konstantinos P. Triantis,
for his unwavering faith in me during my entire research. He continuously motivated me
and helped me in learning to think qualitatively and holistically, as opposed to my
unifocal mathematical and physical view of the world.
I would also like to thank Dr. William G. Sullivan and Dr. Charles P. Koelling who
offered their useful inputs and helpful advice as my research committee members.
My parents have been hugely instrumental in whatever I have been able to accomplish till
today. Their love and support at all times has seen me through good times and bad, more
than I can put in words.
I would like to thank Robert Schatzel and Kirt Kerr (Newport News Shipbuilding), and
Irv Chewning, Jimmy Moy, Nicole Ray, and Anne Hasson (Naval Sea Command Cost
Estimating Group) for their invaluable contributions in the group modeling process. I
v
would also like to thank the Office of Naval Research for their funding support of this
research project.
My research colleagues Wassana Siangdung and John Scott helped me with my research
work a number of times and I am thankful to them for their support.
Pavinder Monga
vi
Table of Contents
1. Introduction ............................................................................................................. 1
1.1 Introduction to the Problem .................................................................................. 1
1.2 Research Objectives................................................................................................ 2
1.2.1 System Dynamics Modeling .............................................................................. 3
1.2.2 Performance Drivers/ Dynamic Hypotheses...................................................... 3
1.2.3 Dynamic Behavior ............................................................................................. 4
1.3 Motivation for Research......................................................................................... 5
1.4 Methodology Overview........................................................................................... 7
1.4.1 Problem Articulation.......................................................................................... 7
1.4.2 Formulating Dynamic Hypotheses .................................................................... 8
1.4.3 Formulating a Simulation Model ....................................................................... 9
1.4.4 Testing and Validation....................................................................................... 9
1.4.5 Policy Design and Evaluation............................................................................ 9
1.5 Overview of the Results ........................................................................................ 10
1.6 Organization of the Thesis ................................................................................... 11
2. Literature Review ................................................................................................ 12
2.1 System Dynamics .................................................................................................. 12
2.1.1 Origins and Fundamental Notions of System Dynamics ................................. 12
vii
2.1.2 SD Behaviors ................................................................................................... 14
2.1.3 Causal Loop Diagrams..................................................................................... 16
2.1.4 Stocks and Flows ............................................................................................. 17
2.2 Knowledge Elicitation and Group Modeling...................................................... 20
2.3 Real World versus Virtual World ....................................................................... 25
2.4 New Technology Implementation: Development and Integration ................... 27
2.5 Performance Metrics ............................................................................................ 33
3. The Model ............................................................................................................... 36
3.1 Problem Articulation ............................................................................................ 36
3.1.1 The Problem..................................................................................................... 36
3.1.2 Two Powerful Tools in the Initial Characterization of the Problem................ 37
3.2 Formulating Dynamic Hypotheses ...................................................................... 39
3.2.1 Endogenous Explanation ................................................................................. 39
3.2.2 The system ...................................................................................................... 41
3.3 Variable Definitions .............................................................................................. 44
3.4 The Causal Loop Diagram ................................................................................... 50
3.4.1 Causal Loop Diagram Notation ....................................................................... 51
3.4.2 Technology Development Causal Loop Diagram............................................ 51
3.5 The Quantitative Description: Formulating a Simulation Model .................... 56
3.5.1 Stocks and Flows ............................................................................................. 57
3.5.2 TD Funding Stock and Flow Structure ............................................................ 61
viii
3.5.3 Technology Development Effort Stock and Flow Structure............................ 63
3.5.4 TD Testing Effort Stock and Flow Structure................................................... 66
3.5.5 TD Actual Testing Results Stock and Flow Structure ..................................... 67
3.5.6 TD Management Effort Stock and Flow Structure .......................................... 70
3.5.7 TD Actual Costs Stock and Flow Structure..................................................... 74
3.5.8 TD Cost Overrun Fraction Structure................................................................ 75
3.5.9 TD Training Imparted Stock and Flow Structure ............................................ 77
3.5.10 TD Risk Structure .......................................................................................... 78
4. Results, Testing, Sensitivity Analysis, Validation and Verification... 82
4.1 Results .................................................................................................................... 83
4.1.1 Simulation Control Parameters ........................................................................ 83
4.1.2 User defined parameters .................................................................................. 83
4.1.3 Technology Development Effort ..................................................................... 88
4.1.4 TD Testing Effort............................................................................................. 89
4.1.5 TD Management Effort.................................................................................... 90
4.1.6 TD Actual Costs............................................................................................... 91
4.1.7 TD Redevelopment, TD Results Discrepancy, and Actual Testing Results.... 94
4.2 Hypotheses Testing: Cost Performance Drivers ................................................ 96
4.3 Sensitivity Analysis ............................................................................................. 106
4.4 Testing, Verification, and Validation ................................................................ 109
4.4.1 Face Validity.................................................................................................. 110
4.4.2 Structure Assessment Tests............................................................................ 111
ix
4.4.3 Dimensional Consistency Tests ..................................................................... 112
4.4.4 Integration Error Tests ................................................................................... 112
4.4.5 Behavior Reproduction Tests......................................................................... 115
5. Conclusion ............................................................................................................ 117
5.1 Overview of the Results ...................................................................................... 117
5.2 Verification of the Dynamic Hypotheses........................................................... 117
5.3 Policy Suggestions ............................................................................................... 119
5.4 Future Issues........................................................................................................ 120
References................................................................................................................... 123
Appendix A. Glossary ........................................................................................... 127
Appendix B. System-subsystem structure ..................................................... 130
Appendix C. Evolution of Causal Loop Diagrams ..................................... 134
x
List of Figures
Figure 2.1 System dynamics structures and their behavior (Sterman, 2000) .................. 15
Figure 2.2 Example of a Causal Link .............................................................................. 17
Figure 2.3 General Structure of a Stock and Flow........................................................... 18
Figure 2.4 The Interaction of the Virtual and Real World (ONR proposal, 2000).......... 27
Figure 2.5 Conceptual structure of Cooper’s model........................................................ 31
Figure 2.6 Conceptual structure of Abdel-Hamid and Madnick’s model........................ 33
Figure 3.1 Reference modes for key variables................................................................. 38
Figure 3.2 Example of a causal link................................................................................. 50
Figure 3.3 Technology Development Causal Loop Diagram .......................................... 52
Figure 3.4 General Structure of a Stock and Flow........................................................... 58
Figure 3.5 Technology Development stock and flow diagram........................................ 60
Figure 3.6 TD Funding Stock and Flow Structure........................................................... 61
Figure 3.7 Impact of Funding Stability on TD Funding Inflow Rate .............................. 62
Figure 3.8 Technology Development Effort Stock and Flow Structure .......................... 63
Figure 3.9 The Impact of TD Risk on the rate of Technology Development.................. 65
Figure 3.10 TD Testing Effort Stock and Flow Structure ............................................... 66
Figure 3.11 TD Actual Testing Results Stock and Flow Structure.................................. 67
Figure 3.12 Impact of TD Results Discrepancy on TD Redevelopment Fraction........... 70
Figure 3.13 TD Management Effort Stock and Flow Structure....................................... 71
Figure 3.14 Impact of Funding Stability on Management Initiation Rate....................... 72
Figure 3.15 TD Actual Costs Stock and Flow Structure ................................................. 74
Figure 3.16 TD Cost Overrun Fraction Structure ............................................................ 75
xi
Figure 3.17 TD Training Imparted Stock and Flow Structure......................................... 77
Figure 3.18 TD Risk Structure......................................................................................... 78
Figure 3.19 Impact of TD Percentage Training on TD Risk ........................................... 79
Figure 3.20 Impact of Complexity of New Technology on TD Risk .............................. 80
Figure 3.21 Impact of Technology Maturity on TD Risk................................................ 81
Figure 4.1 Feedback structure causing damped oscillation ............................................. 86
Figure 4.2 Technology Development Stock and Flows Behavior ................................... 88
Figure 4.3 TD Testing Stock and Flows Behavior .......................................................... 89
Figure 4.4 TD Management Stock and Flows Behavior.................................................. 90
Figure 4.5 TD Actual Costs and Cost Realization Rate................................................... 92
Figure 4.6 Feedback structure causing goal seeking ....................................................... 93
Figure 4.7 TD Redevelopment Fraction and Results Discrepancy.................................. 94
Figure 4.8 TD Results Enhancement Rate and Actual Testing Results........................... 95
Figure 4.9 TD Costs at three Training levels................................................................... 97
Figure 4.10 TD Cost Overrun Fraction at three training levels ....................................... 98
Figure 4.11 Impact of TD Results Discrepancy on TD Redevelopment Fraction........... 99
Figure 4.12 TD Costs at three redevelopment profiles .................................................. 100
Figure 4.13 TD Cost Overrun Fraction at three training levels ..................................... 101
Figure 4.14 TD Results Enhancement Rate and Actual Results at three redevelopment profiles ............................................................................................................................ 102
Figure 4.15 TD Costs at three levels of complexity of technology ............................... 103
Figure 4.16 TD Costs at three levels of technology maturity ........................................ 105
Figure 4.17 Technology Development at three sets of parameter inputs ...................... 107
Figure 4.18 TD Costs at three sets of parameter inputs................................................. 108
Figure 4.19 TD Costs for a simulation run with the integration time step=0.125 week 113
xii
Figure 4.20 TD Costs for a simulation run with the integration time step=0.03125 week......................................................................................................................................... 114
Figure 4.21 TD Costs for a simulation run with the integration time step=0.0078125 week ................................................................................................................................ 114
Figure 4.22 TD Cost Realization Rate........................................................................... 115
Figure 4.23 TD Actual Testing Results ......................................................................... 116
1
Chapter 1. Introduction
1.1 Introduction to the Problem
In today’s world, the fact that technology is all-pervasive is well known and realized.
Sophisticated and rapidly changing technology is the foundation for a vast majority of
products and services we depend on. Nevertheless, although technology is everywhere,
its development and real-world applications are still faced with tremendous problems for
technology users as well as technology developers and implementers.
The introduction of new technologies leads to increases in costs due to unforeseen system
performance degradation, additional downtime, and increased maintenance over a
system’s life cycle. Within this context, performance refers to the specific measures
related to the technical competence or operational capability of the technology. It can be
thought of as the degree to which the system reflects (meets or exceeds) operational
requirements. Cost overruns form part of one of the important control mechanisms in
implementation processes, wherein cost overruns are traded off against technical
performance realizations. The traditional systems engineering implementation process
can then be thought of as the chief reason for the ineffective development of new
technologies. In the traditional implementation of the systems engineering process as far
as the management of research and development, emphasis is usually placed on breaking
the various activities of the process into discrete and non-dynamic process phases that are
isolated in structure and function (Roberts, 1964). This is different from how the process
really works, wherein the different stages in the technology development process actually
“talk” to each other on a continuous basis. The technology development process consists
2
of upfront research, design, engineering and development, prototype development (or
procurement), testing of prototypes (development and non-development), and
adaptability studies (impact on interfacing systems and operations) (definition provided
by Newport News Shipbuilding and Naval Sea Command Cost Estimating Group experts,
and discussed in detail in Chapter 3, Section 3.2.2). These different pieces of the
technology development process are closely interrelated with some of the activities
occurring later. This provides a control feedback to earlier occurring activities; thus
giving a dynamic nature to the whole process.
1.2 Research Objectives
The three main objectives in the research are identified as:
• To build a System Dynamics performance assessment framework model for the
technology development process.
• To identify the performance drivers in the technology development process.
• To investigate and understand the dynamic behavior that characterizes the
technology development process.
These objectives are pertinent to some of the future challenges in the System Dynamics
field. System Dynamics has been applied to various fields in the natural and social
sciences. There still remain countless problems and issues where understanding is
lacking and the dominant theories are event-oriented rather than dynamic in nature
(Sterman, 2000). Within the realm of System Dynamics modeling, understanding the
connection between System Dynamics model structure and model behavior in complex
3
model formulations is a big challenge (Richardson, 1996). Profound understanding
comes after a prolonged series of model tests of deepening sophistication and insight.
1.2.1 System Dynamics Modeling
The traditional system engineering process addresses the management of technology
development by breaking it into discrete parts such as “preliminary design”, “detailed
design” and “hardware” (Roberts, 1964). These parts are assumed to be independent and
isolated. The parts are still further broken down into subordinate tasks, giving the overall
management of technology development a “work breakdown structure” look. There are
two primary drawbacks to such an approach. The first is that the dynamic nature of the
research and development process is completely ignored. The emphasis is placed upon
discrete sets of events, separated in time and lacking any base understanding of the
underlying common elements that bind them (Roberts, 1964). The second drawback is
that the fundamental systems nature of research and development is ignored. Research
and development is essentially an iterative process and is driven by an “action-results-
information-new action” methodology. This is disregarded to a large extent in the
implementation of the traditional system engineering process. Thus, there is a
compelling need to study the dynamics of the process. This research attempts to do that
by means of using System Dynamics.
1.2.2 Performance Drivers/ Dynamic Hypotheses
The chief performance metrics (as tracked by Newport News Shipbuilding) in the
Technology Development process are cost, risk, schedule, and technical performance.
4
Newport News Shipbuilding, located and headquartered in Newport News, Virginia, is
one of the top ten defense companies in the United States. Newport News Shipbuilding
designs, builds, and maintains nuclear powered aircraft carriers and submarines for the
U.S. Navy, and also provides maintenance and repair services for a wide variety of
military and commercial ships.
Cost is defined as the amount of dollars expended in the various development efforts like
project management, design, engineering, and testing and evaluation. Risk is measured
as the overall measure of probability of failure of the occurrence of specific desired
events. Schedule is defined as the time taken to complete the entire research and
development process. Technical performance refers to the specific performance
characteristics related to the operational capability of the technology. This research
hypothesizes that performance is driven by the dynamic nature of the management of the
technology development process. The implementation of the traditional systems
engineering approach to the problem fails to “see” the overall integrated picture, making
local decisions, which though being good for the specific task at hand, adversely affect
overall performance. In contrast, the system dynamics approach takes into account the
inherent dynamic nature of the technology development process.
1.2.3 Dynamic Behavior
The fundamental modes of observed behavior in dynamic systems are exponential
growth, goal seeking, and oscillation (Sterman, 2000). These modes of behavior are
discussed in detail in Chapter 2, Section 2.1.2. Exponential Growth arises from self-
reinforcing feedback. The greater a quantity is, the greater is its net change (increase or
5
decrease), and this feedback to the process further augments the net change. Goal
Seeking arises from self-controlling feedback. Within this context, negative feedback
loops tend to oppose any changes or deviations in the state of the system; they tend to
restore equilibrium and hence are goal seeking. Oscillation arises due to negative
feedback with significant time delays. Due to time delays in the effects of the actions,
corrective action taken to restore the equilibrium state or to achieve the goal of a system
continues even after the equilibrium has been reached. Thus the goal is overshot.
Corrective action is taken again to correct for the overshot value. One of the objectives
of the research is to identify and study the dominant behavior prevailing the dynamics of
the technology development process. This behavior could be one of the fundamental
modes of dynamic behavior or could conceivably be an interaction of two or more of the
dynamic behaviors explained above.
1.3 Motivation for Research
One of the chief motivations for this research was to better understand the system
engineering process as it pertains to the technology development process within the scope
of new technology implementation in complex systems or organizations. As mentioned
earlier, the implementation of the traditional system engineering process fails to address
the dynamic nature of the technology development process. Not much work has been
done in the past to study the management of technology development from a “dynamics”
perspective.
6
The second motivating factor for this research was the impact of the System Dynamics
model on its end users. The Office of Naval Research funded this research and the Navy
is going to be the end user of an overall cost-estimating model, of which the System
Dynamics model of the technology development process is a part. According to experts
from NAVSEA (Naval Sea Systems Command) Cost Estimating Group, Newport News
Shipbuilding, and the Office of Naval Research (all of who have taken part during the
current group modeling process described later in this Chapter), the Navy has been
repeatedly experiencing cost and schedule overruns in large new technology
implementation projects on aircraft carriers. Given the current atmosphere of fast-
changing technologies, a highest-level system mission for the Navy is to keep upgrading
and/or replacing obsolete technology to maintain the highest levels of Mission
Preparedness and Condition Readiness (see Glossary, Appendix A). Thus, an objective
for the Navy is to better manage the new technology implementation projects in terms of
achieving reduced cost and schedule overruns. A certain degree of buy-in to the concept
of System Dynamics exists in the Navy. The System Dynamics modeling methodology
has been used earlier by Cooper (1980) to model the management of a large ship design
and construction program so as to describe its dynamic structure, and as a result it
quantified the causes for cost overruns in that program. Additionally since 1996, the
Navy has funded the development of the Operations and Support Cost Analysis Model
(OSCAM) for its operations and support activities. This is a system dynamics model as
well.
7
1.4 Methodology Overview
Modeling of any system is fundamentally a creative and intensive process. Assumptions
are made at various steps of the modeling process. These assumptions need to be tested
from the data that are gathered and analyzed from the field, and the models are then
revised based on the results. However, there are no particular strictly defined rules of
modeling (Sterman, 2000). The involvement of the decision-makers at all the steps of
modeling process is very crucial. Their views and active participation are very important
for successful and meaningful modeling.
Overview of the Modeling Process
1.4.1 Problem Articulation
The identification of a clear purpose is a very important step. Models are most effective
when designed for a small problem/part of the system rather than the whole system itself.
Identification of a clear purpose based on a problem in the system helps in making
decisions about the framing of the model: what it should include and what should be left
out.
Two powerful processes in the initial characterization of the problem are the framing of
the reference modes and the definition of the time horizon.
Reference Modes: These are graphs and other descriptive data showing the development
of the problem over time. These graphs help the modelers and the decision-makers break
free from the narrow event-oriented outlook to a broader system-wide outlook. The
reference modes help characterize the problem dynamically, that is, as a pattern of
behavior, unfolding over time, which shows how the problem arose and how it might
8
evolve in the future. To develop reference modes, it is important to first identify a time
horizon for the problem and to define the variables and concepts that are considered
important for understanding the problem.
Time Horizon: This is a very important feature in any problem characterization, wherein
the time to study the problem is chosen.
1.4.2 Formulating Dynamic Hypotheses
The next step in the modeling process is to develop a working theory that explains the
problem at hand. Since the theory should account for the dynamics of the behavior of the
system based on the underlying feedbacks and interactions between its different
components, it is called a dynamic hypothesis. The role of the modeler is to elicit the
views of the decision-makers involved with the problem. System Dynamics seeks
endogenous explanations for phenomena rather than exogenous ones. Explanations
based on exogenous (see Glossary (Appendix A)) entities/variables are not of much use
as they articulate dynamics of endogenous (see Glossary (Appendix A)) variables in
terms of exogenous variables whose behavior was assumed in the first place and cannot
be changed. Therefore, an Endogenous Explanation for the system in terms of the
interactions and feedback relationships among different components/variables is
proposed. Mapping System Structure and a Model Boundary Chart Development (see
Glossary (Appendix A)) are important tools used in Step 2; they communicate the
boundary of the system and list the key endogenous and exogenous variables of the
system respectively. All main subsystems are represented in a Subsystem Diagram and
9
the various inputs, outputs, and constraints on the individual subsystems are also drawn.
The links/interactions between the different subsystems are also represented.
1.4.3 Formulating a Simulation Model
This next step in the modeling process involves setting up a formal model complete with
equations, parameters and initial conditions that represent the system. Given the
complexity of systems, real-world experiments are often impractical and infeasible.
Therefore, a simulation model needs to be developed in the virtual realm.
1.4.4 Testing and Validation
This step involves the testing of the model as to whether it replicates the behavior of the
real-world system. Another important task of testing is to verify whether the variables
and parameters of the model have a meaningful concept in the real world. When models
are subjected to extreme conditions, their robustness is determined. A model should not
just be a means to mirror field data exactly. In addition, elements of the model should
have a sound conceptual basis for being included.
1.4.5 Policy Design and Evaluation
Model-based policy analyses involve the use of the model to help investigate why
particular policies have the effects they do and to identify policies that can be changed to
improve the problematic behavior of the real system. New policies can be formulated
and their impact on system performance under alternative scenarios can be evaluated.
This is done once the model has been developed and has gained the modeler’s and
10
decision-maker’s confidence. Interactions between different policies can be complex in
many cases and thus need to be properly taken into account.
1.5 Overview of the Results
The System Dynamics Technology Development model was developed and simulated.
Newport News Shipbuilding (NNS) and Naval Sea Command (NAVSEA) Cost
Estimating Group experts provided the user-input parameters. The model per se is not
technology specific as it is mostly process specific. This means that it is generalizable.
The results obtained from running the simulation are discussed in detail in Chapter 4.
The simulation was run for two years (104 weeks), the time horizon of the technology
development process. The simulation results showed two main modes of dynamic
behavior. One dynamic behavior was the damped oscillation observed for these
variables: Technology Development (TD) Effort, TD Testing Effort, TD Management
Effort, and TD Actual Costs Realization rate. This was attributed to the presence of
oscillatory structures (in the overall causal loop and stock and flow structures) that are
characterized by a set of negative feedback loops (presented in Figure 4.1). The other
dynamic behavior observed was the goal seeking observed for the variable Actual Testing
Results. The feedback structure causing this type of dynamic behavior is identified in
Figure 4.6. The dynamic hypotheses (discussed in detail in Chapter 3, Section 3.2.1)
were tested using the model developed in VENSIM Professional 4.0 by varying
parameters and observing the changes in the subsequent results from the simulation.
Some sensitivity analysis was also performed on the model. Sensitivity analysis was
done using three sets of key parameter combinations, namely, (1) very high technology
11
complexity-very immature technology-no training, (2) medium technology complexity-
medium mature technology-average training, and (3) very low technology complexity-
very mature technology-high training. These were chosen to represent two extreme
condition scenarios and an average condition scenario.
1.6 Organization of the Thesis
The thesis is organized in five chapters, including the first chapter which is an
introduction to the research. Chapter 2 is a literature review in the fields of System
Dynamics, Knowledge Elicitation and Group Modeling, and New Technology
Development and Integration. The System Dynamics model and the modeling process
are presented in Chapter 3. The results obtained from the model and the testing,
sensitivity analysis, validation and verification completed are presented in Chapter 4.
Chapter 5 concludes with an overview of the results, some policy suggestions, and a
discussion on the future research areas.
12
Chapter 2. Literature Review
2.1 System Dynamics
“The system approach is the modus operandi of dealing with
complex systems. It is holistic in scope, creative in manner,
and rational in execution. Thus, it is based on looking at a total
activity, project, design, or system, rather than considering the
efficiency of the component tasks independently. It is
innovative, in that rather than seeking modifications of older
solutions to similar problems, new problem definitions are
sought, new alternative solutions generated, and new measures
of evaluation are employed if necessary” (Drew, 1995, pp. 4).
2.1.1 Origins and Fundamental Notions of System Dynamics
System Dynamics (SD) is a policy modeling methodology based on the foundations of
(1) decision making, (2) feedback mechanism analysis, and (3) simulation. Decision-
making focuses on how actions are to be taken by decision-makers. Feedback deals with
the way information generated provides insights into decision-making and effects
decision-making in similar cases in the future. Simulation provides decision-makers with
a tool to work in a virtual environment where they can view and analyze the effects of
their decisions in the future, unlike in a real social system.
13
Forrester first used the concept of System Dynamics in an article entitled “Industrial
Dynamics: A Major Breakthrough for Decision Makers” which appeared in Harvard
Business Review in 1958. His initial work focused on analyzing and simulating micro-
level industrial systems such as production, distribution, order handling, inventory
control, and advertising. Forrester expanded his system dynamics techniques in
Principles of Systems in 1968, where he detailed the basic concepts of system dynamics
in a more technical form, outlining the mathematical theory of feedback system dynamics
(Forrester, 1968).
A stark feature of modern times is continuous change. Changes in existing elements
often encounter resistance from people themselves (for whose betterment the changes
were sought in the first place). The problems we face today are often too complex and
dynamic in nature, i.e., there are many factors and forces in play that we do not
comprehend easily. Also, these factors and forces themselves are very dynamic in nature.
Systems thinking is advocated by many “thinkers” who advocate holistic thinking and the
conceptualization of “systems” wherein “everything is connected to everything else”.
Systems Dynamics is an approach whose main purpose is to understand and model
complex and dynamic systems. It employs concepts of nonlinear dynamics and feedback
control, concepts that will be discussed in detail shortly.
Feedback
Actions taken on an element in a system result in changes in the state of the element.
These, in turn, bring about changes in other linked elements, and the effects may trail
14
back to the “first” element. This is called feedback. Feedbacks are of two types: 1)
Positive or self-reinforcing, which amplify the current change in the system; and 2)
Negative or self-correcting, which seek balance and provide equilibrium by opposing the
change taking place in the system. Complex systems are “complex” because of the
multiple feedbacks/interactions among the various components of the system.
2.1.2 SD Behaviors
The feedback structure of a system generates its behavior. Most dynamics observed in
systems fall under three fundamental modes of behavior: exponential growth, goal
seeking, and oscillation (Sterman, 2000). These modes of behavior are shown in Figure
2.1. Exponential Growth arises from positive or self-reinforcing feedback. The greater a
quantity is, the greater is its net change (increase/decrease), and this is the feedback to the
process that further augments the net change. Thus this is a self-reinforcing feedback and
there is an exponential growth/decline. Goal seeking behavior arises from negative or
self-controlling feedback. Negative feedback loops tend to oppose any changes or
deviations in the state of the system; they tend to restore equilibrium and hence are goal
seeking. The rate of change diminishes as the goal is approached, such that there is a
smooth attainment of the goal/equilibrium state of the system. Oscillation arises due to
negative feedback with significant time delays. Corrective action to restore an
equilibrium state or to achieve the goal of the system continues even after the equilibrium
has been reached due to time delays in identifying the effects of the actions on the
system. Thus the goal is overshot. Corrective action taken again (negative feedback
loop) leads to undershooting and hence oscillation. The principle that behavior is a result
15
Figure 2.1 System Dynamics Structures and their Behavior (Sterman, 2000)
T
State of theSystem
State of theSystem + System Net
Increase
+
+
T
State of theSystem
Goal
State of theSystem
CorrectiveAction
Discrepancy
System Goal-
++
-+
T
State of theSystem
Goal
State of theSystem
Correc tiveAction
Discrepancy
System Goa l -
++
-+
Delay
Delay
Delay
T
State of theSystem
SystemProductionCapacity
State of theSystem
+System NetIncrease
SystemProductionCapacity
Discrepancy+Net Increase
+
-
+
++-
Consumptionof ProductionCapacity
+
-
Goal SeekingExponential Growth
Overshoot and Collapse
Oscillation
T
State of theSystem
System Production Capacity
State of theSystem
+System NetIncrease
SystemProductionCapacity
Discrepancy+
Net Increase
+
-
+
++-
S-shaped Growth S-shaped Growth with Overshoot
System Production Capacity
T
State of theSystem
State of theSystem+System Net
Increase
SystemProductionCapacity
Discrepancy+Net Increase
+
-
+
++-
Delay
16
of the structure of the system enables a discovery of the system structure (its feedback
loops, non-linear interactions) by observing the behavior of the system. Therefore, when
the pattern of behavior is observed, conclusions can be drawn about the dominant
feedback mechanisms acting in the system.
Nonlinear interactions among the three major feedback structures give rise to other
complex patterns of behavior of the systems (Sterman, 2000). S-shaped growth arises
when there is a positive feedback initially, and later negative feedback dominates, leading
to attainment of equilibrium by the system. S-shaped growth with overshoot occurs
when, after an initial exponential growth phase, negative feedback with time delays kicks
in. In this case, the system oscillates around the equilibrium state. Overshoot and
collapse occurs as a result of the equilibrium state itself declining after the exponential
growth phase has commenced, and negative feedback is triggered. Since the equilibrium
declines, a second negative feedback gets activated, wherein the system approaches the
new equilibrium state.
2.1.3 Causal Loop Diagrams
The feedback structure of complex systems is qualitatively mapped using causal
diagrams. A Causal Loop Diagram (CLD) consists of variables connected by causal
links, shown by arrows. Each link has a polarity. A positive (denoted by “+” on the
arrow) link implies that if the cause increases (decreases), the effect increases (decreases)
above (below) what it would otherwise have been. A negative (denoted by “-” on the
17
arrow) link implies that if the cause increases (decreases), the effect decreases (increases)
below (above) what it would otherwise have been (Sterman, 2000).
For example,
Variable 1Birth Rate Population
+
Causal Link Link Polarity
Variable 2
Figure 2.2 Example of a Causal Link
Causal loops are immensely helpful in eliciting and capturing the mental models of the
decision-makers in a qualitative fashion. Interviews and conversations with people who
are a part of the system are important sources of quantitative as well as qualitative data
required in modeling. Views and information from people involved at different levels of
the system are elicited, and from these, the modeler is able to form a causal structure of
the system.
2.1.4 Stocks and Flows
Causal loops are used effectively at the start of a modeling project to capture mental
models. However, one of the most important limitations of the causal diagrams is their
inability to capture the stock and flow structure of systems. Stocks and flows, along with
feedback, are the two central concepts of dynamic systems theory. Stocks are
accumulations as a result of a difference in input and output flow rates to a
process/component in a system. Stocks give the systems inertia and memory, based on
18
which decisions and actions are taken. Stocks also create delays in a system and generate
disequilibria (Sterman, 2000).
Notation:
All stock and flow structures are composed of stocks (represented by rectangles), inflows
(represented by arrows pointing into the stock), outflows (represented by arrows pointing
out from the stock), valves, and sources and sinks for flows (represented by clouds).
Stock
General Structure
Inflow Outflow
Figure 2.3 General Structure of a Stock and Flow
Mathematical Representation of Stocks and Flows
Stock (t) = Stock (t0) + ∫ [Inflow(t) – Outflow (t)] dt
Stocks are the state variables or integrals in the system. They accumulate (integrate) their
inflows less their outflows. Flows are all those which are rates or derivatives. If a
snapshot of a system was taken at any instant of time, what would be seen is the state of
different processes or components of the system. These are the stocks in the systems.
The inflows and outflows are what have been frozen and so cannot be identified. Stock
19
and flow networks undoubtedly follow the laws of conservation of material. The
contents of the stock and flow networks are conserved in the sense that items entering a
stock remain there until they flow out. When an item flows from one stock to another,
the first stock loses exactly as much as the second one gains.
Auxiliary variables
Auxiliary variables are often introduced in stock and flow structures to provide a better
understanding. Auxiliary variables are neither stocks nor flows; they are functions of
stocks and exogenous inputs (see Glossary (Appendix A)). They are variables used for
computational convenience.
The contribution of Stocks to Dynamics is multifold: (1) Stocks denote the state of a
particular element in the system, and based on this information, decisions can be made or
actions can be taken. (2) They provide the system with inertia and memory. For
example, intangible stocks like beliefs and memories characterize our mental states. (3)
They induce delays in the system. A stock or accumulation occurs when the output lags
the input to a process, and whenever this happens, delays occur. (4) Stocks decouple
inflow from outflow. Inflows and outflows are controlled or decided upon by different
people/resources in the system. A difference in inflow and outflow rates creates
disequilibria.
20
2.2 Knowledge Elicitation and Group Modeling
The topic of group discussion and decision-making is of significant interest vis-à-vis SD
model building as system dynamics modelers have done intensively interactive modeling
with decision-makers for a long time. SD modelers typically rely on multiple, diverse
streams of information to create and calibrate model structure. According to Vennix et
al. (1992), the most productive source of information in these streams is the information
contained within the mental models of the key actors in a system, and accessing the
minds of these experts and actors in a system is largely an art. The academic preparation
of SD modelers rarely includes formal training to help build formal skills in eliciting
information for model building.
Some formal work in the field of model building process and types of tasks to be carried
out with decision-making groups has been done. Richardson and Pugh (1981) defined
seven stages in building a SD model: problem identification and definition, system
conceptualization, model formulation, analysis of model behavior, model evaluation,
policy analysis, and model use or implementation. Roberts et al. (1983) proposed an
almost identical set of six steps to organize their model-building approach. Problem
identification and definitions includes going through the steps of formally defining the
problem, identifying a time horizon, defining the level of abstraction and aggregation,
and defining the system boundaries. It also includes what should be included within the
purview of the system and what should not. System conceptualization consists of
establishing the relevant variables in the system, mapping relationships between the
variables, determining important causal loop feedback structures, and generating dynamic
21
hypotheses as proposed explanations of the problem. Model formulation entails
developing mathematical equations and quantifying the model parameters. In the
analysis or evaluation stage, the model is checked for logical values and sensitivity
analyses are conducted. Policy analysis includes conducting experiments on the model
by changing key policies and observing and analyzing the resulting changes in the model
outputs.
Eliciting information residing in the mental models of the decision-makers of the system
is facilitated by two important techniques: divergent thinking and convergent thinking
(Vennix et al., 1992). Divergent thinking implies different key persons of the system
coming up with their opinions or thoughts on issues, as an individual exercise. It is most
helpful and meaningful in the problem definition and model conceptualization phases
where an individual or a group is attempting to determine what factors or variables to
include or exclude from a system’s boundary. One of the approaches to facilitate
divergent thinking in the modeling process is called the “Nominal Group Technique”
approach, developed by Andre P. Delbecq and Andrew H. Van de Ven in 1968, and first
tested in 1969. It began as a technique to enhance the effectiveness and efficiency of
program planning in health services. It is now used extensively in areas of productivity
measurement systems development, strategic planning, and strategy implementation.
Sink (1983) delineated a five-staged structured process through which the Nominal
Group Technique helps groups generate ideas and reach consensus: (1) Individual silent
generation of ideas, (2) Individual round-robin feedback from group members of their
22
ideas, which are recorded on a flip chart, (3) Group clarification of each recorded idea,
(4) Individual voting and ranking on priority ideas, and (5) Discussion on group
consensus results and focus on potential next steps. The last of the above stages
essentially is the “convergent thinking” exercise, wherein the group works together on
the ideas it has come up with, to build a consensus on approach and various issues.
Although much experience has been gained in the area of modeling as learning, it seems
that building models with decision-making groups is still more art than science. In recent
years, significant interest and efforts have been concentrated on coming up with a more
formal approach to the craft of group model building. Andersen, Richardson, and Vennix
(1997) talk about “adding more science” to the craft of model building in SD. One of the
important elements they discuss are the various goals of group model building: (1)
Mental model alignment of the various key people involved in the system, (2) Creating
agreement (consensus) about a policy or decision, and (3) Generating commitment with a
particular decision.
In their efforts to formalize the modeling process, Andersen and Richardson (1997)
proposed a set of scripts for group model building. They use the term “scripts” to
represent small behavioral descriptions of facilitated group exercises that move a group
forward in a systems thinking process or activity. The expected results at the end of a
modeling group activity are: (1) A precise description of the problem to be solved; (2) A
stakeholder analysis in which the key players (both the top management personnel and
the modeling team who will execute the model) of the target organization are identified;
23
and (3) A sketch of the model structure. In order to obtain these results, they delineate a
series of “scripts”.
PLANNING FOR THE GROUP MODELING MEETING
The planning phase for the group modeling sessions involves making a detailed and
careful script of the entire process. The main issues in building the scripts are:
1) Goal Setting:
Interviews with key managers or “gatekeepers” are conducted to gain an initial
understanding of the problem. A gatekeeper is a contact person within the target
organization who helps select the appropriate people within the organization, helps plan
the modeling sessions, and works closely with the modeling group. The audience role
and purpose needs to be clarified by identifying the key stakeholders (preferably from top
management/decision makers), the experts in the technology, and the internal modeling
team. The session products or the expected deliverables at the end of the session need to
be clarified.
2) Logistics:
An appropriate room layout for the modeling session needs to be planned, where every
member can see everyone else. Richardson and Andersen (1995) advise assigning two to
five roles in the session to individuals, such as facilitator, modeler, process coach,
recorder, and gatekeeper. A facilitator conducts and manages the group session, with an
aim of facilitating the expression of views of all the key players involved in the modeling
exercise. A modeler’s role is to be a keen observer in the modeling exercise by listening
and capturing the ideas and views of different session members and to provide inputs on
24
the technical modeling aspects as and when required. A process coach ensures that the
modeling group does not deviate substantially from the modeling agenda, and provides
feedback to the facilitator in terms of the effectiveness of the modeling process being
followed. A recorder takes detailed notes on the various session activities and
developments. A gatekeeper is a contact person within the target organization who helps
select the appropriate people within the organization, helps plan the modeling sessions,
and works closely with the modeling group.
3) Types of Group Task Structures:
During the modeling session, tasks can vary from divergent (brainstorming) tasks to
ranking and evaluating tasks, to integrative or design-oriented tasks. Divergent tasks are
ones in which different key personnel of the target organization come up with their
opinions or thoughts on issues about the system and the problem as an individual
exercise. In convergent tasks, the group works on the ideas it has come up with to build a
consensus on an approach and on various issues. The group also ranks and evaluates
different emerging ideas by voting.
Ford and Sterman (1998) additionally provide a structured method for knowledge
elicitation from the decision-makers. The first step involves the creating of an
environment in which elicitation will occur. This is achieved by a brief description of the
model purpose, the major subsystems, the roles of each subsystem and the relationships
to be characterized. The next step is to focus on one relationship at a time. The
facilitator describes the relationship operationally by identifying the input and output
variables (with units of measure) which the relationship describes. The next step is a
25
visual description whereby the experts visualize and imagine the flow of inputs and
outputs in the relationship. The purpose of this step is to activate, bound, and clarify the
experts’ mental images of the relationship (Ford and Sterman, 1998). The next step in
the elicitation process is to obtain a graphical description of the relationship by means of
graphical plots of the output variables versus the input variables in the relationship. The
next step is to have a group discussion on the relationships as characterized by different
experts. The aim of such a discussion is to better understand the relationship and not so
much as to resolve the differences in the different characterizations of the relationship by
different experts.
2.3 Real World versus Virtual World
Throughout this research, the relationship between the real and virtual world has been
constantly kept in mind. A modeling and simulation effort should have the primary goal
of establishing policies or procedures that can be implemented in the real world.
However, many modelers and performance measurement teams often lose sight of the
real world implementation. It was important for everyone involved with this research to
understand the relationship between the real and virtual worlds and the strengths and
weaknesses of each. Figure 2.4 shows the interactions between the virtual and the real
world.
In the virtual world the system structure is known and controllable. Information is
complete, accurate and the feedback is almost instantaneous from actions within the
26
system. Decisions and policy implementations are usually perfect. The goal of the
virtual world is learning (Sterman, 1994).
The goal of real world systems is performance. In the real world, systems often have
unknown structures, long time delays, and experiments are uncontrollable. Given these
factors, often policies that are implemented into the real world system do not realize their
full impact until long after the policy is implemented. Information from real world
systems is often incomplete due to long delays, and data is often biased or contains errors
(Sterman, 1994).
The erroneous or incomplete data often leads to inaccurate mental models. Mental
models serve as the basis for one’s beliefs of how a system works. Receiving inaccurate
or incomplete data about the system often leads to erroneous beliefs of how the system
performs (Sterman, 1994). This condition leads to the establishments of rules, system
structure, or strategies that govern the system in a sub-optimal manner. When strategies,
policies, structure, or decision rules are inaccurate, decisions made within the system are
also incorrect. This creates negative consequences for the system processes and leads to
a vicious circle for declining system performance. However, modeling offers the
decision-maker with the opportunity to learn about system behavior, ask pertinent
questions, collect accurate information and define more effective policies. The constant
and effective interface between the virtual and real world allows for better decision-
making over the long run.
27
Figure 2.4 The Interaction of the Virtual and Real World (Triantis and Vaneman, 2000)
Today, given the complex systems within which new technologies are implemented,
effective incorporation and implementation of a new technology poses a formidable
challenge in almost all industries and organizations. There exists a vast body of literature
on the research efforts in new technology implementation processes and issues.
2.4 New Technology Implementation: Development and Integration
“ ‘They did not anticipate that the steel axe would lead to more
sleep, prostitution, and a breakdown of social relationships and
customs.
ProblemDefinition
SystemConceptualization
ModelFormulation
Optimization
PolicyAnalysis
ModelBehavior and
Validation
Mental Models ofReal World
Strategy, Structure,Decision Rules
Decisions(Organizational
Experiments) InformationFeedback
Actions
REAL WORLD SYSTEM
VIRTUAL WORLD SYSTEM
DynamicHypothesis
28
Change agents frequently do not sense or understand the social
meaning of the innovations that they introduce.’
Rogers (1995), speaking from Sharp (1952: 69-92)
The preceding quotation illustrates the possible consequences when implementers do not
anticipate long-term and side effects of a change. The implementers of the steel axe did
not foresee the consequences of their introduction of a relatively simple technology.
These missionaries believed that the aboriginals would use steel axes as they had their
stone ones (Rogers, 1995). The missionaries did not anticipate that this new technology
would evoke new understandings of the technology. Today, given the complex systems
within which new technologies are implemented, it is even more difficult for
implementers of modern technologies (e.g., flexible manufacturing systems, customizable
voice and electronic mail systems, biotechnology) to anticipate users’ understanding of
the technology and the side-effects of the technology (e.g., Weick, 1990). ”
-- quoted in part from Griffith (1999, pp. 473).
Effective incorporation and implementation of a new technology poses a formidable
challenge in almost all industries and organizations. Often new technologies hold
tremendous promise for enhancing operational efficiency and effectiveness in
organizations. Much of this potential, however, is never realized, much more often due
to poor technology management rather than technical shortcomings. According to
Griffith et al. (1999), a major cause of failure of technology innovations is the inability of
organizations to develop effective implementation processes. Thus, project managers
29
have the responsibility to create a concept of how implementation funding, technology
integration, and support are interrelated.
Most of the work in literature on new technology implementation is focused on
implementation issues in manufacturing processes. Most of the research is oriented on
human-issues, which essentially means that it seeks to explore how humans react at a
social level to the implementation of new technology in their work environment. Most of
the literature focuses on whether certain variables such as participation of the topmost
level managers makes a difference, and does not focus on the nature of the underlying
process of technology implementation. Goodman and Griffith (1991) explored new
technology implementation with a process approach view. However, their study focuses
on manufacturing organizations. They identify five critical processes in technology
implementation, namely, socialization, commitment, reward allocation,
feedback/redesign, and diffusion. Socialization refers to the processes by which
individuals acquire skills about the new technology. Commitment refers to the binding of
the individual in the organization to certain behavioral acts relevant to the technology.
This may be in the form of educating the workers about the long-term benefits of the
technology and inculcating in them a sense of loyalty for need for the implementation of
the new technology. Reward allocation refers to the allocation of different types of
benefits and incentives to workers who are associated with the new technology
implementation. Data are collected once the technology is in place and in the transient
operating phase. The transient operating phase of a new technology is the operating
phase after the technology has been introduced in the organization and before it has
30
achieved its full working potential. Based on feedback, redesign activities are undertaken
to enhance the operation of the new technology. Most new technologies follow the path
of evolution in structure, process, and outcomes. This means that as the technology
spreads throughout the organization, one would expect the creation of a social
environment for the emergence of value consensus on the new technology. The
technology spread, or so-called “diffusion,” also signals its legitimacy and acceptance
into the overall structural fabric of the organization.
All literature cited to this point looks at technology as Commercial-Off-The-Shelf
products that are implemented in complex organizations. Almost no literature exists that
explores the development process of a technology and then its integration into a complex
system. The development process of a technology consists of various research and
development (R&D) activities. The process includes project management, research,
requirements definition, specification development, engineering, modeling and
simulation, drawing development, hardware and software development, system
architecture development, and testing (Iansiti, 1997). There is a lack of a system level
understanding of the dynamics of the technology development process, both from
management and process perspectives.
There are some references in the literature on work done in the area of the SD approach
applied to project management issues. The SD approach to project management is based
on a holistic view of the project management process and focuses on the feedback
processes that take place within the project system. Cooper (1980) developed a SD
31
model at Pugh-Roberts Associates that was the first major practical application of System
Dynamics to project management. It was used to quantify the causes of cost overruns in
a large military shipbuilding project. Further versions of the model were developed and
used to support a strategic analysis of prospective shipbuilding programs. One of the
major novelties of this work was the concept of the rework cycle, a structure at the core
of the model that explicitly incorporates the concepts of undiscovered rework, time to
discover rework, work quality, and varying staff productivity.
Manpower
Availability ofprior work
Subsequent workprogress
Undetected errors inprior work
Discovery of priorwork errors
Productivity
Prior work quality
Figure 2.5 Conceptual Structure of Cooper’s (1980) model
One of the key relationship structures of Cooper’s model is shown in Figure 2.5. The
structure centers around the quality of work performed. Cooper contended that out-of-
sequence, incomplete, and/or incorrect work has serious impacts on the performance of
subsequent work. The need for rework (because work is discovered to be incomplete or
incorrect) may delay or impair dependent work in subsequent phases. The need for
32
rework may remain undiscovered until progress in a subsequent phase becomes directly
dependent on the required work.
Cooper contended that cost overruns in large Navy ship construction projects could be
broken into two majors segments: 1) Overruns as a result of the direct impact of a design
change, or the “hard-core” costs, and 2) Overruns due to “delay and disruption” costs –
the second and third order “ripple effects” of dealing with the direct changes. These
snowballing effects are the most difficult to quantify and justify. Cooper’s model was
successful in describing the dynamic behavior of such rippling effects and capturing the
dynamic structure of the shipbuilding process that leads to the cost overruns attributed to
the indirect effects resulting from direct design changes.
Some studies have examined the dynamics of specific types of projects. Abdel-Hamid
and Madnick (1991) developed a SD model for the evaluation of a software development
project. Figure 2.6 below shows the structure of the software production process
proposed by them.
According to Abdel-Hamid and Madnick, the software development lifecycle includes
designing, coding and testing phases. They contend that as the software is developed, it
is also reviewed to detect any errors, e.g., using quality assurance activities such as
structured walkthroughs of the software code. Errors detected through such activities are
reworked. They further propose that not all software errors are detected during
development; some “escape” detection until the testing phase.
33
SoftwareDeveloped
Software Tested
Errors ThatEscaped
Errors Detected andCorrected in Testing
Phase
Errors
Errors Detected andCorrected in Development
Phase
Software DevelopmentProductivity
Manpower
Software TestingProductivity
Quality AssuranceEffort
Figure 2.6 Conceptual Structure of Abdel-Hamid and Madnick’s (1991) Model
The decision-makers in a technology development process evaluate the process based on
certain performance measures that are tracked during the entire life cycle. The chief
performance metrics are discussed in the next section.
2.5 Performance Metrics
The key performance metrics of interest are cost, schedule, technical performance, and
risk.
34
Cost: Cost is viewed as the total life cycle costs (LCC) of the technology implementation.
It would include costs starting from the R&D phase going all the way up to the disposal
of the technology.
Cost is the amount paid or payable for the acquisition of materials, property, or services
(NNS and NAVSEA experts). It is also used with descriptive adjectives such as
"acquisition cost", or "product cost," etc. Although dollars normally are used as the unit
of measure, the broad definition of cost equates to economic resources, i.e., manpower,
equipment, real facilities, supplies, and all other resources necessary for weapon, project,
program, or agency support systems and activities. (Source: Defense Acquisition Desk
Book, Parametric Cost Estimating Handbook)
Schedule: Schedule refers to the time elapsed for a particular task. The time elapsed is
both actual versus the planned time. Schedule includes program events/activities, their
inter-relationships defined by program phase and activity/event time durations. The
events may represent management objectives and provide for management and control of
such effort (NNS and NAVSEA experts).
Risk: Risk is the probability of failure of a new technology. Risk addresses the problems
or issues associated with technical, cost, and scheduling aspects of designing, testing,
manufacturing, operating and supporting, and disposing of processes, systems and
equipment. Risks affect the ability to successfully achieve cost, schedule, and technical
performance objectives. Potential sources of risk include new processes like new
designs, new operational requirements, immature or emerging technologies, new
35
performance requirements, and political or organizational changes (NNS and NAVSEA
experts).
Technical Performance: It is the degree to which the system reflects (meets or exceeds)
the expected operational characteristics.
Performance is a statement of requirements in terms of required results with criteria for
verifying compliance, without stating methods for achieving the required results. A
performance specification defines the functional requirements for the item, the
environment in which it must operate, and the interface and interchangeability
requirements (NNS and NAVSEA experts).
36
Chapter 3. The Model
The modeling process used for the development of the model followed the steps and
guidelines presented by Sterman in “Business Dynamics: Systems Thinking and
Modeling for a Complex World” (2000). This process is summarized subsequently.
3.1 Problem Articulation
The most important step in modeling is the problem articulation (Sterman, 2000). The
identification of a clear purpose is very critical. Models are effective when designed for a
small problem or part of the system rather than the whole system itself. Identification of
a clear purpose based on a problem in the system helps in making decisions about the
framing of the model, i.e., what it should include and what should be left out.
3.1.1 The Problem
Costs and schedule overruns are commonplace in large research and development
projects. Cost overruns are exhibited usually when there is a need to hire and train
additional personnel midway through the project. Schedule overruns are experienced
when allotted time is not met. A point to be noted is that not all research and
development projects have these problems. However, we proceeded with the assumption
that they have persisted in our client’s (Newport News Shipbuilding) experiences in spite
of reasonable attempts to avoid them. We considered here a large technology
development project, involving a large number of people, a considerable number of
detailed tasks, and a long time frame (104 weeks).
37
3.1.2 Two Powerful Tools in the Initial Characterization of the Problem
Reference Modes: These are a set of graphs and other descriptive data showing the
development of the problem over time (Sterman, 2000). These graphs assist the decision-
makers and the modelers to break free from the narrow event-oriented perspective to a
broader wholistic perspective and to events that are removed in time and space.
The behavior pattern of certain key variables was elicited from the decision-makers.
Those variables were:
Risk: It is defined as the probability of failure of the new technology. Risk addresses the
problems or issues associated with technical, cost, and scheduling aspects of designing,
testing, manufacturing, operating and supporting, and disposing of processes, systems
and equipment. Risks affect the ability to successfully achieve cost schedule and
technical performance objectives (Group Modeling exercise, as theorized by Richardson
and Pugh (1981), Roberts et al. (1983), Vennix et al. (1992), and Andersen et al. (1997),
and discussed in Chapter 2, Section 2.2).
Cost: It is viewed as the total life cycle costs (LCC) of the technology implementation. It
would include costs starting from the R&D technology development phase going all the
way through the disposal of the technology. Cost is the dollar amount paid for the
acquisition of materials, property, and services. In contract and proposal document
language, it denotes dollars and amounts exclusive of fee or profit (Group Modeling
exercise, as theorized by Richardson and Pugh (1981), Roberts et al. (1983), Vennix et al.
(1992), and Andersen et al. (1997), and discussed in Chapter 2, Section 2.2).
38
Performance: It is the degree to which the system reflects (meets or exceeds) the
expected operational characteristics. Performance is a statement of requirements in terms
of required results with criteria for verifying compliance, without stating methods for
achieving the required results. A performance specification defines the functional
requirements for the item, the environment in which it must operate, and the interface and
interchangeability requirements (Group Modeling exercise, as theorized by Richardson
and Pugh (1981), Roberts et al. (1983), Vennix et al. (1992), and Andersen et al. (1997),
and discussed in Chapter 2, Section 2.2).
Examples of reference modes for these variables are depicted in Figure 3.1 below.
$ Figure 3.1 Reference modes for Key Variables
Time (Years) 1 2
Cost
Technical Performance
39
Time Horizon: This is a very important feature in any problem characterization. Most
conventional approaches focus on studying the problem and the system over a short time
horizon. This is mainly due to our event-oriented outlook. We need to realize that the
problem might have originated a long time back, and also that actions taken now may
cause effects that are far displaced in time and space. Therefore, selecting an adequately
long time horizon is very important. It was assumed that the technology development
process takes two years to complete. Therefore, the time horizon of the technology
development process was defined to be two years (104 weeks).
3.2 Formulating Dynamic Hypotheses
This is the next step in modeling in which a working theory is developed to explain the
problem at hand. Since the theory should account for the dynamics of the behavior of the
system based on the underlying feedbacks and interactions between its different
components, it is called a dynamic hypothesis. The role of the modeler is to elicit the
views of the decision-makers that can affect the problem in the system, and to develop
hypotheses to explain the problem.
3.2.1 Endogenous Explanation
System Dynamics seeks endogenous explanations for phenomena rather than exogenous
ones. Explanations based on exogenous (see Glossary (Appendix A)) entities/variables
are not of much interest as they articulate dynamics of endogenous (see Glossary
(Appendix A)) variables in terms of other variables whose behavior is assumed (Sterman,
2000). Exogenous variables are “outside” the system and therefore they are not affected
40
dynamically by interactions and feedbacks among the variables/entities within the
system. Hence, an explanation of the system behavior in terms of these exogenous
variables cannot capture the dynamics of the system. There should be very few
exogenous variables in the model of the system under study. A broad boundary should
be determined for the system as opposed to a narrow limiting boundary.
The following hypotheses are proposed. Running simulations on the Technology
Development model can test all the four hypotheses.
1) A lack of appropriate training causes cost overruns in the technology development
process. The right people may not be available to accomplish the tasks necessary
in technology development. Training is one of the important issues in the
execution of any large-scale project. New people recruited have to be trained to
get to the level of knowledge that experienced people have. In the case of
technology development projects, training becomes all the more important given
that technicians, engineers, and workers have to be adequately trained to get
familiar with the new technology.
2) Rework in a project adds to the cost overruns. The notion of rework refers to the
fact that not all work done in the course of a large project is flawless. Some
fraction of it is less than satisfactory and must be redone. Unsatisfactory work is
not found out right away, however. For some time it passes off as real progress,
until the need for reworking the tasks involved resurfaces (see the model
presented by Cooper (1980), Chapter 2, Section 2.4).
41
3) An increase in the complexity (see definition in Section 3.2.2 of this chapter) of
the new technology causes an increase in the total costs incurred. As technology
becomes more complex, it requires more effort to be put in technology
development. As a result, total costs increase.
4) An increase in the maturity (see definition in Section 3.2.2 of this chapter) of the
technology decreases the total costs incurred.
3.2.2 The system
The system was defined as the process that is responsible for the technology
development, within the scope of the implementation of new technologies, on aircraft-
carriers.
(Please see Appendix B for a system/subsystem diagram and brief descriptions of the
various subsystems.)
In the development phase, new technologies are developed and/or assessed. The various
technology development activities are research and development (R&D) activities that
include project management, research, requirements definition, specification
development, engineering, modeling and simulation, drawing development, hardware and
software development, system architecture development, and testing.
42
Risk is another significant variable in this system, and results from the inherent difficulty
in developing and/or implementing new technology. Regardless of whether the new
technology is being developed, or is already developed and needs to be implemented, the
technology risks need to be assessed for their impact on existing/future ship systems and
operations.
Broadly, technology development has two main subtasks:
1) The development of the technology, and
2) The assessment of the new technology.
Technology development involves the following activities (based on NNS and Naval Sea
Systems Command Cost Estimating group (NAVSEA) experts).
1) Up-front research and development: This involves research activities based on
available literature and/or innovations and alterations to the existing technology
designs available in the literature. Various alternatives are studied in the
technology field to gain an initial insight as to which alternative would best meet
the operational requirements.
2) Prototype development: Based on the chosen technology alternative, this activity
involves the development of a prototype or miniature physical model that
provides for the desired process/project requirements and objectives.
3) Testing of prototypes: In this activity, the developed prototypes are tested for
compliance with all desired requirements. Testing is fundamentally concerned
43
with an assessment of how well the technology performs with respect to technical
requirements/aspects.
4) Adaptability studies: These studies are conducted to assess the adaptability of the
technology within the operating environment of the ship. This means how well
the technology can ease into the existing operations on the ship without causing
much disruption to other processes that are not directly related to the operation for
which the technology is being developed.
5) Development of drawing specifications: Once the prototype has been developed
and tested favorably as a first iteration, the drawing specifications are developed
for the full-size unit of the technology product. Drawing specifications are a
detailed list of the physical components, characteristics, and dimensions of the
technological unit to be fabricated.
6) Development of component drawings: Based on the drawing specifications of the
previous stage, this activity involves the development of detailed drawings for the
various components of the technological unit.
7) Incorporate flexibility as it relates to the life cycle of the new technology: One of
the important issues in the successful implementation of new technologies is to
account for the fact that technologies might become obsolete quicker than their
planned/expected life-term. This is very important in the case of electronic and
computer hardware and software technologies. The technology development
process needs to address flexibility and plan for the possible technology upgrades
and replacements before the expected or anticipated life-term of the technological
unit.
44
8) The design of the infrastructure support processes: New technology
implementation on ships is fraught with the possibility that the depot services
might lack the infrastructure support material and equipment for maintenance of
the new technology. The technology development process should plan for the
anticipated support requirements and design the same if they are non-existent in
the current depot facilities.
All the above activities except testing were aggregated as one activity, namely,
technology development, in the model. The aggregation level was as desired by the
experts.
3.3 Variable Definitions
The variables of the system were identified as a part of the group modeling exercise as
presented by Richardson and Pugh (1981), Roberts et al. (1983), Vennix et al. (1992),
and Andersen et al. (1997), and discussed in Chapter 2, Section 2.2. Each participant in
the group modeling process was asked to list the key variables she or he felt were
important in the system. A consensus was then built on a common list of variables for
the system. A similar exercise was undertaken to obtain the definitions of the identified
variables.
45
Definitions of the Identified Variables in the Technology Development Process
1) Funding – It is the amount of money provided (from sources outside the system
and allocated by the program management) to carry out all the work required in
the system. The variable is measured using cost (units are Year 2001 dollars).
2) Technology Maturity – It is defined as the state of development of a new
technology, i.e., how mature it is. Increased technology maturity implies reduced
development or procurement risk, because the technology is better defined. It is a
dimensionless variable measured using a relative scale (varying from 1 to 5;
1=very immature, 2=immature, 3=medium maturity, 4=mature, and 5=very
mature).
3) Technology Development – This is the effort (including both labor and materials)
required to develop the new technology. It is a measure of the “largeness” of the
development activity. Technology Development includes all activities except
testing. It is measured using human effort time (units are man-hours).
4) Actual Testing Results – These are the actual results produced from testing the
technology being developed. Testing results are measured with respect to the
technical performance issues related to the new technology. It is a dimensionless
variable measured using a relative scale (varying from 0 to 1; 0 corresponds to
total failure and 1 corresponds to total success).
46
5) Target Testing Results – These are the desired results from testing the technology
being developed. They are drawn from the specifications or high-level
requirements of the operational characteristics of the new technology. It is a
dimensionless variable measured using a relative scale (varying from 0 to 1; 0
corresponds to total failure and 1 corresponds to total success, model-user
specified it as 1).
6) Integration Risk – It is a measure of the risk involved in integrating the
technology onboard ship. This is separate from the risk involved in actually
developing the technology. Integration Risk is a dimensionless variable measured
using a relative scale (varying from 1 to 10; increasing from a very low risk at 1
to a very high risk at 10).
7) Technology Development Technical Performance – This is a measure of how
favorable the technology being developed is, i.e., how likely it is to be successful
for the intended application. This variable is a direct translation of the final
Actual Testing Results obtained at the end of the Technology Development phase,
i.e., at the end of two years. It is a dimensionless variable measured using a
relative scale (varying from 0 to 1; 0 corresponds to total failure and 1
corresponds to total success).
47
8) Technology Development Risk – It is defined as a measure of the risk involved in
developing the new technology. This is separate from the risk involved in
integrating the new technology onboard the ship. It is a dimensionless variable
measured using a relative scale (varying from 1 to 10; increasing from a very low
risk at 1 to a very high risk at 10).
9) Complexity of New Technology – This is a measure of how complex the new
technology is. Increased complexity implies increased technology development
effort. It is a dimensionless variable measured using a relative scale (varying
from 1 to 5; 1=very low complexity, 2=low complexity, 3=medium complexity,
4=high complexity, 5=very high complexity).
10) Technology Development Management – This is a measure of the effort required
to manage and administratively support the technology development process, the
testing process, and the management of the cost overruns. It is measured using
human effort time (units are man-hours).
11) Technology Development Redevelopment fraction – This is the amount of
redevelopment that has to be done to narrow the discrepancy between the actual
and target testing results. It is a dimensionless variable measured as a decimal
value between 0 and 1.
48
12) Technology Development Results Discrepancy – It is the difference between the
actual and target testing results. It is a dimensionless variable measured as a
decimal value between 0 and 1.
13) Training – It is defined as the amount of training imparted to the technical
professionals in the technology development process. It is measured using cost
(units are Year 2001 dollars).
14) Testing Effort – It is the effort (including both labor and materials) required to test
the new technology. It is a function of the amount of work done in technology
development. It is measured using human effort time (units are man-hours).
15) Actual Costs – It is the cumulative costs associated with the technology
development, testing effort, and technology development management effort. It is
measured using cost (units are Year 2001 dollars).
16) Cost Overrun fraction – It is defined as the amount by which the actual costs
exceed the allotted funding (as a fraction of the allotted funding) for the various
activities in the technology development phase. It is a dimensionless variable
measured as a decimal value greater than or equal to 0.
17) Funding Stability – It is defined as a measure of how stable the external funding
source is. It is a dimensionless variable measured using a relative scale (varying
49
from 1 to 10; increasing from a very low stability at 1 to a very high stability at
10).
Assumptions of the model
The following assumptions of the model were obtained from the group modeling
sessions.
1) Funding stability has a positive effect on the amount of funding received for the
technology development phase. A high funding stability ensures a steady flow of
funding whereas a low stability means a dwindling of the funding rate.
2) A low funding stability calls for an increased effort in project management, as the
management has to perform a juggling act to make things happen within a
funding constraint.
3) A cost overrun compels a reduction in the planned technology development
activity so as to try to meet the budget constraints.
4) The more complex a technology is, the more is the anticipated technology
development effort.
5) A redevelopment effort drives up the actual testing results obtained in the testing
process. The assumption is that redevelopment is carried out with an aim to
reduce the discrepancy between the actual and target testing results, and it results
in an increase in the actual state (testing results) of the system.
6) The more mature the technology is, the lesser is the technology integration (on the
ship) risk, the lesser is the amount of effort required for technology development,
and the lesser is the technology development risk.
50
7) An increase in the actual technical performance of the technology drives down the
risk associated with integrating the technology on board the ship.
8) Complexity of technology, funding, funding stability, target testing results, and
technology maturity are exogenous variables to this system. They are not
influenced directly by any other variables from within the system. The user of the
model defines these parameters.
3.4 The Causal Loop Diagram
The feedback structure of complex systems is qualitatively mapped using causal
diagrams. A Causal Loop Diagram (CLD) consists of variables connected by causal
links, shown by arrows. Each link has a polarity. A positive (denoted by “+” on the
arrow) link implies that if the cause increases (decreases), the effect increases (decreases)
above (below) what it would otherwise have been. A negative (denoted by “-” on the
arrow) link implies that if the cause increases (decreases), the effect decreases (increases)
below (above) what it would otherwise have been (Sterman, 2000).
For example:
Variable 1Birth Rate Population
+
Causal Link Link Polarity
Variable 2
Figure 3.2 Example of a Causal Link
51
3.4.1 Causal Loop Diagram Notation
Positive linkages are presented with blue colored arrows and negative linkages are
presented with red colored arrows. The main causal loops identified for the technology
development system (depicted by Figure 3.3) are as follows:
COM-R: Cost Overrun-Management reinforcing loop
DCO-B: Development-Cost Overrun balancing loop
TCO-B: Testing-Cost Overrun balancing loop
MCOD-B: Management-Cost Overrun-Development balancing loop
MCODT-B: Management-Cost Overrun-Development-Testing balancing loop
DTrR-B: Development-Training-Risk balancing loop
ReTDis-B: Redevelopment-Testing Results-Discrepancy balancing loop
The Causal Loop Diagram was obtained from the group modeling sessions. Figure 3.3 is
the final version of the causal loop diagram. Previous versions are included in Appendix
C.
3.4.2 Technology Development Causal Loop Diagram
The overall causal loop diagram of the system is shown in Figure 3.3.
Description of the Technology Development Causal Loop Diagram
The various feedback loops manifested in the causal loop diagram are described in detail
herewith.
52
TechnologyMaturity
TechnologyDevelopment Risk
Complexity of newTechnology
TechnologyDevelopmentManagement
TechnologyDevelopment
Integration Risk
Funding
Testing Effort
Redevelopment
TechnologyDevelopmentPerformance
Funding Stability
Actual TestingResults
+
+
+
-
+
Target TestingResults
Discrepancy
-
++
-
-
+
+
+
+
+
Training
-
Cost Overrun Actual Costs+
-
-
+
+ +
+
COM-R
DCO-B TC
O-B
MCODT-B
MCOD-B
ReTDis-B
+
+
+
Figure 3.3 Integrated Technology Development Causal Loop Diagram
Cost Overrun-Management Reinforcing Loop (COM-R)
Cost Overruns are defined as the amount by which the actual costs exceed the available
funding during the technology development phase of the new technology implementation
process. As cost overruns increase (decrease), the management group within the
technology development phase has to increase (decrease) its management effort
53
associated with juggling the available funds among various activities occurring in the
phase. Or in other words, the financial management effort increases (decreases) as the
cost overruns increase (decrease). Now the management effort itself is expressed as the
man-hours required to complete the actual effort. So, as the management effort increases
(decreases), the actual costs in the technology development phase go up (down), thus
further driving up (down) the cost overruns. This loop behaves as a reinforcing loop.
Development-Cost Overrun Balancing Loop (DCO-B)
As technology development effort goes up (down), the man-hours required to carry out
the actual effort goes up (down). This drives up (down) the actual costs of the
technology development phase. As the costs increase (decrease), the cost overruns
increase (decrease). An increase (decrease) in the cost overruns fuels a decrease
(increase) in the technology development activity as a control feedback mechanism. This
loop behaves as a balancing loop.
Testing-Cost Overrun Balancing Loop (TCO-B)
When the technology development effort increases (decreases), it leads to an increase
(decrease) in the testing effort associated with the amount of technology development
taking place. An increase (decrease) in testing effort implies an increase (decrease) in the
man-hours associated with the actual effort exerted. This drives up (down) the actual
costs of the technology development phase, in turn driving up (down) the cost overruns.
An increase (decrease) in the cost overruns fuels a decrease (increase) in the technology
54
development activity as a control feedback mechanism. This loop behaves as a balancing
loop.
Management-Cost Overrun-Development Balancing Loop (MCOD-B)
As the technology development effort increases (decreases), the accompanying
management effort associated with the management process of the technology
development activity increases (decreases) as well. The implied increase (decrease) in
man-hours associated with the actual effort results in an increase (decrease) in the actual
costs of the technology development phase, in turn driving up (down) the cost overruns.
An increase (decrease) in the cost overruns fuels a decrease (increase) in the technology
development activity as a control feedback mechanism. This loop behaves as a balancing
loop.
Management-Cost Overrun-Development-Testing Balancing Loop (MCODT-B)
When Technology development effort increases (decreases), it leads to an increase
(decrease) in the testing effort associated with the amount of technology development
taking place. This in turn results in an increase (decrease) in the technology development
management activity to manage the testing effort. An increase (decrease) in management
effort implies an increase (decrease) in the man-hours associated with the actual effort
put in. This drives up (down) the actual costs of the technology development phase, in
turn driving up (down) the cost overruns. An increase (decrease) in the cost overruns
fuels a decrease (increase) in the technology development activity as a control feedback
mechanism. This loop behaves as a balancing loop.
55
Redevelopment-Testing Results-Discrepancy Balancing Loop (ReTDis-B)
An increase (decrease) in the redevelopment activity leads to an increase (decrease) in the
actual state of the system. This means that the testing results get better and closer to the
target testing results. An increase (decrease) in actual testing results leads to a decrease
(increase) in the discrepancy from the target testing results. As this discrepancy
decreases (increases), the redevelopment rate also decreases (increases), as there is a
lower (higher) rate need for development improvement to achieve the target performance.
This loop behaves as a balancing loop. This loop can be used to test the second
hypothesis stated earlier in this chapter.
An increase (decrease) in funding stability decreases (increases) the amount of financial
management effort associated with juggling the funding obtained. An increase (decrease)
in funding causes an increase (decrease) in the technology development effort. It also
leads to a decrease (increase) in the cost overruns.
It should be noted here that there is no feedback loop in the model that will test the third
and fourth dynamic hypotheses of the model. These hypotheses are imbedded in the
relationship defined between complexity/maturity and technology development risk
(Equation 3.24). Furthermore, there is no direct linkage between the integration risk and
the development effort or any other activity in this subsystem. However, this notion of
integration risk is used in the technology integration and operation, support, and disposal
subsystems and impacts activities in these subsystems.
56
An increase (decrease) in complexity of the new technology leads to an increase
(decrease) in the technology development effort. An increase (decrease) in technology
maturity results in a decrease (increase) in the risk associated with technology
development, which consequently leads to a decrease (increase) in the technology
development effort and the risk associated with integration of the technology onboard the
ship.
An increase (decrease) in the redevelopment rate increases (decreases) the technology
development effort. An increase (decrease) in the actual state of the system that is
reflected in the actual testing results, leads to an increase (decrease) in the overall
technology development technical performance. This increase (decrease) in the overall
technology development technical performance causes a decrease (increase) in the risk
associated with the integration of the technology onboard the ship. However, as noted
earlier, the integration risk variable is not part of a feedback mechanism in this subsystem
but affects activities in the technology integration and operations, support, and disposal
subsystems.
3.5 The Quantitative Description: Formulating a Simulation Model
This next step in modeling involves setting up a formal model complete with equations,
parameters and initial conditions that represent the system. Sometimes the dynamic
hypotheses can be tested directly through data collection or experiments in the real
system. However, in many situations, real-world experiments are often impractical and
57
infeasible due to a number of reasons. To be able to test dynamic hypotheses by
conducting experiments in the real system, one would have to conduct the experiments
several times over with different starting parameters or with changes in the different
interactions between the entities of the system. The costs to conduct several experiments
in the real system are often prohibitive. In addition, if the testing of a dynamic
hypothesis involves conducting several experiments with some different parameters but
in the same time frame, it is infeasible. Therefore, a simulation model needs to be
developed in the virtual realm. The involvement of the decision-makers at all these steps
of the modeling process is very crucial. Their views and active participation in the
problem definition, hypothesis formulation, and simulation modeling are very important
for successful and meaningful modeling.
In the group modeling sessions, the equations describing the relationships between the
various variables were elicited from the clients. They were asked for their inputs on the
units for measurement of different variables, the functional form of the various equations
between variables, parameters of these equations (elicited through graphical portrayal of
key relationships), and the initial values of all stock variables.
3.5.1 Stocks and Flows
Causal loops are used effectively at the start of a modeling project to capture mental
models. However, one of the most important limitations of the causal diagrams is their
inability to capture stock and flow structure of systems. Stocks and flows, along with
feedback, are the two central concepts of dynamic systems theory. Stocks are
58
accumulations that occur as a result of a difference in input and output flow rates to a
process/component in a system. Stocks give the systems inertia and memory, based on
which decisions and actions are taken. Stocks also create delays in a system and generate
disequilibria (Sterman, 2000).
Stock and Flow Notation:
All stock and flow structures are composed of stocks (represented by rectangles), inflows
(represented by arrows pointing into the stock), outflows (represented by arrows pointing
out from the stock), valves, and sources and sinks for flows (represented by clouds).
Stock
General Structure
Inflow Outflow
Figure 3.4 General Structure of a Stock and Flow
Mathematical Representation of Stocks and Flows
Stock (t) = Stock (t0) + ∫ [Inflow(t) – Outflow (t)] dt (3.1)
Stocks are the state variables or integrals in the system. They accumulate (integrate) their
inflows less their outflows. Flows are all the variables that are rates or derivatives. Were
a snapshot of a system to be taken at any instant of time, what would be seen is the state
59
of different processes or components of the system. These are the stocks in the systems.
The inflows and outflows are what have been frozen and so cannot be identified. Stock
and flow networks undoubtedly follow the laws of conservation of material. The
contents of the stock and flow networks are conserved in the sense that items entering a
stock remain there until they flow out. When an item flows from one stock to another,
the first stock loses exactly as much as the second one gains. Stocks follow the
mechanism of physical flow wherein items in a stock physically move to another stock,
thus depleting the first stock and augmenting the second.
Stock and Flow Diagram Notation
The technology development stock and flow diagram is represented by Figure 3.5. In this
figure, TD represents Technology Development.
The stock and flow structure has a one-to-one correspondence to the causal loop structure
presented earlier. A causal loop diagram is a graphical tool to qualitatively capture the
mental models of the decision-makers. A stock and flow diagram is a more detailed
graphical tool to help quantify what has already been captured in the causal loop diagram.
Variables and concepts in the causal loop diagram are manifested as stock and flow
structures in the stock and flow diagram. There are eight stock-and-flow structures in the
stock and flow diagram and they are discussed herewith.
60
Technology Development Stock and Flow Diagram
TechnologyDevelopment
EffortTechnologyDevelopment Initiation
Rate
TechnologyDevelopment
Completion Rate
Complexity of NewTechnology
TDTestingEffortTD Testing
Initiation RateTD Testing
Completion Rate
TDActualCostsTD Cost
Realization Rate+
TDManagement
EffortTD ManagementInitiation Rate
TD ManagementCompletion Rate
+ +
+
TDFundingTD Funding
Inflow RateTD Funding
Allocation Rate
+
Funding Stability
+
-
AccumulatedTD FundingAccumulated TD
Funding Inflow Rate
+
TD Cost OverrunFraction
-
+
-
TD TrainingImparted
TD Training Rate
TD Risk
+
TechnologyMaturity
Integration Risk
-
TD ActualTesting Results
TD ResultsEnhancement Rate
TD RedevelopmentFraction
TD ResultsDiscrepancy
TD Target TestingResults
-
+
+
+
+
TD TechnicalPerformance
-+
DCO-B
TCO-B
+
+
MCOD-B
MCODT-B
COM-R
DTrR-B
ReTDis-B
Manhour CostRate
Dollar PerWeek
+
+
TD PercentageTraining
+
-Initial TD Risk+
+
+
TD Testing toDevelopment Fraction
TD TestingResidence Time
TD Management toDevelopment Fraction
TD Management toTesting Fraction
TD ManagementResidence Time
Technology DevelopmentEffort Residence Time
TD FundingResidence Time
+
Figure 3.5 Technology Development Stock and Flow Diagram
61
3.5.2 TD Funding Stock and Flow Structure
TDFundingTD Funding
Inflow RateTD Funding
Allocation Rate
Funding Stability
+
Dollar PerWeek
TD FundingResidence Time
Figure 3.6 TD Funding Stock and Flow Structure
The TD Funding stock (dollars) is fed into by a TD Funding Inflow Rate (dollars/week)
and is depleted by a TD Funding Allocation Rate (dollars/week). The TD Funding stock
variable is an integral of the TD Funding Inflow Rate less the TD Funding Allocation
Rate.
TD Funding (t) = TD Funding (0) + ∫ [TD Funding Inflow Rate
– TD Funding Allocation Rate] dt (3.2)
TD Funding (0) = 0
The TD Funding Inflow Rate has a base value of $25000/week (NNS and NAVSEA
experts, and is obtained from the program management subsystem). An increase
(decrease) in Funding Stability leads to an increase (decrease) in the TD Funding Inflow
Rate. Funding Stability (on a scale of 1 to 10; with 1 being most unstable, increasing in
62
stability up to 10 being the most stable) effects the base value of TD Funding Inflow Rate
by a divisive factor. When Funding Stability is 10, the divisive factor is 1. When
Funding Stability is 4, the divisive factor is 1.1. When Funding Stability is 1, the divisive
factor is 1.3. A quadratic curve (of the form y = ax2+bx+c) was fitted through the three
points to obtain an analytical expression capturing the effect of Funding Stability on the
TD Funding Inflow Rate. A quadratic curve was assumed based on the visual shape of
the curve elicited from the clients for the effect of Funding Stability on the TD Funding
Inflow Rate. The curve fitting was done manually. The graph depicted below (Figure
3.7) was plotted in MS Excel 2000.
1
1.1
1.3 y = 0.0056x2 - 0.0944x + 1.3889
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1 2 3 4 5 6 7 8 9 10
Funding Stability
Div
isiv
e Fa
ctor
Figure 3.7 Impact of Funding Stability on TD Funding Inflow Rate
The overall equation (with inputs from NNS, NAVSEA experts) to describe the
relationship is:
TD Funding Inflow Rate = 25000 / (0.0056*Funding Stability2 - 0.0944*Funding
Stability + 1.3889) * Dollar Per Week (3.3)
63
The TD Funding Allocation Rate is a first order delay (see Glossary (Appendix A)) of the
TD Funding stock.
TD Funding Allocation Rate = TD Funding / TD Funding Residence Time (3.4)
The TD Funding Residence Time is the average time the incoming funding stays in the
stock before being allocated.
3.5.3 Technology Development Effort Stock and Flow Structure
TechnologyDevelopment
EffortTechnologyDevelopment Initiation
Rate
TechnologyDevelopment
Completion Rate
TD FundingAllocation Rate
+
TD Cost OverrunFraction
-
TD Risk
+
Manhour CostRate
TD RedevelopmentFraction
+Technology Development
Effort Residence Time
Figure 3.8 Technology Development Effort Stock and Flow Structure
The Technology Development Effort stock (man-hours) is fed into by a Technology
Development Initiation Rate (man-hours/week) and is depleted by a Technology
Development Completion Rate (man-hours/week). The Technology Development Effort
64
stock variable is an integral of the Technology Development Initiation Rate less the
Technology Development Completion Rate.
Technology Development Effort (t) = Technology Development Effort (0)
+ ∫ [Technology Development Initiation Rate
– TD Funding Allocation Rate] dt (3.5)
Technology Development Effort (0) = 0
The Technology Development Initiation Rate is primarily driven by the TD Funding
Allocation Rate. Three-fourths of the funding is allocated for technology development
activities. The Manhour Cost Rate (dollars/man-hour) is the cost of labor and is used to
convert the funding rate units into development rate units (man-hours). The TD
Redevelopment Fraction is a dimensionless variable that captures the amount of
redevelopment done as a fraction of the development activity assigned in the first place.
The amount of redevelopment done adds to the Technology Development Initiation Rate.
The TD Cost Overrun Fraction is a dimensionless variable that captures the amount of
cost overruns over the total allocated funding, as a fraction of the total allocated funding.
It imposes a penalty on the technology development activities by reducing the
Technology Development Initiation Rate at twice the rate of cost overruns. An increase
(decrease) in TD Risk leads to an increase (decrease) in the Technology Development
Initiation Rate. TD Risk (on a scale of 1 to 10; with 1 being very low risk, increasing in
risk up to 10 being very high risk) effects the Technology Development Initiation Rate by
a multiplicative factor. At TD Risk equals 10, the multiplicative factor is 5. At TD Risk
equals 8, the multiplicative factor is 2.2. At TD Risk equals 7, the multiplicative factor is
65
1.8. At TD Risk equals 1, the multiplicative factor is 1. An ellipse curve (of the form
x2/a2 + y2/b2 = 1) was fitted through the four points to obtain an analytical expression
capturing the effect of the TD Risk on the Technology Development Initiation Rate. An
ellipse curve was assumed based on the visual shape of the curve elicited from the
experts for the effect of the TD Risk on the Technology Development Initiation Rate.
The curve fitting was done manually. The graph depicted below (Figure 3.9) was plotted
in MS Excel 2000.
1.82.2
5
1
0
1
2
3
4
5
6
1 2 3 4 5 6 7 8 9 10
TD Risk
Mul
tiplic
ativ
e Fa
ctor /811)-(x-1*4-5y 2=
Figure 3.9 The Impact of TD Risk on the rate of Technology Development
The overall equation (with inputs from NNS, NAVSEA experts) to describe the
relationship is:
Technology Development = TD Funding Allocation Rate / Manhour Cost Rate * 0.75
Initiation Rate * (1+TD Redevelopment Fraction)
* (1-2*TD Cost Overrun Fraction)
* (5-4* 81/)1(1 2−− TDRisk ) (3.6)
66
The Technology Development Completion Rate is a first order delay of the Technology
Development Effort stock.
Technology Development Completion Rate = Technology Development Effort /
Technology Development Effort Residence Time
(3.7)
3.5.4 TD Testing Effort Stock and Flow Structure
TechnologyDevelopment
Completion Rate
TDTestingEffortTD Testing
Initiation RateTD Testing
Completion Rate+
TD Testing toDevelopment Fraction
TD TestingResidence Time
Figure 3.10 TD Testing Effort Stock and Flow Structure
The TD Testing Effort stock (man-hours) is fed into by a TD Testing Initiation Rate
(man-hours/week) and is depleted by a TD Testing Completion Rate (man-hours/week).
The TD Testing Effort stock is an integral of the TD Testing Initiation Rate less the TD
Testing Completion Rate.
TD Testing Effort (t) = TD Testing Effort (0) + ∫ [TD Testing Initiation Rate
– TD Testing Completion Rate] dt (3.8)
67
TD Testing Effort (0) = 0
The TD Testing Initiation rate is primarily driven by the Technology Development
Completion Rate. The amount of testing done (in man-hours/week) is a fraction of the
amount of technology development completed (in man-hours/week).
TD Testing Initiation Rate = Technology Development Completion Rate
* TD Testing to Development Fraction (3.9)
The TD Testing Completion Rate is a first order delay of the TD Testing Effort stock.
TD Testing Completion Rate = TD Testing Effort / TD Testing Residence Time
(3.10)
3.5.5 TD Actual Testing Results Stock and Flow Structure
TD ActualTesting ResultsTD Results
Enhancement Rate
TD RedevelopmentFraction
TD ResultsDiscrepancy
TD Target TestingResults
-
++
+
TD TestingResidence Time
Technology DevelopmentEffort Residence Time
Figure 3.11 TD Actual Testing Results Stock and Flow Structure
68
The TD Actual Testing Results stock (dimensionless) is fed by a TD Results
Enhancement Rate (week-1). The TD Actual Testing Results stock variable is an integral
of the TD Results Enhancement Rate.
TD Actual Testing Results (t) = TD Actual Testing Results (0)
+ ∫ [TD Results Enhancement Rate] dt (3.11)
It is assumed that the technology development activity yields an initial testing results
value of 0.6; i.e., sixth-tenths of the desired goal is met with respect to technology
development when the activity is done for the first time (NNS and NAVSEA experts).
Subsequent redevelopment activity is carried out and it is assumed that redevelopment
enhances the technical performance of the technology developed and hence better testing
results are obtained.
TD Actual Testing Results (0) = 0.6
The TD Results Enhancement Rate is a function of the TD Redevelopment Fraction. As
the TD Redevelopment Fraction increases (decreases), the rate at which testing results get
better increase (decrease). Results get enhanced as redevelopment is done, and this
enhancement occurs over a time span of the technology development delay plus the
testing delay (NNS and NAVSEA experts).
The overall equation (with inputs from NNS and NAVSEA experts) to describe the
relationship is:
TD Results Enhancement Rate = TD Redevelopment Fraction /
(Technology Development Effort Residence Time
+TD Testing Residence Time) (3.12)
69
The TD Results Discrepancy (dimensionless) is the difference between the TD Target
Testing Results (dimensionless, user-defined to be equal to 1) and the TD Actual Testing
Results (dimensionless; varying between 0 and 1).
TD Results Discrepancy = TD Target Testing Results-TD Actual Testing Results
(3.13)
The TD Redevelopment Fraction is a dimensionless variable that captures the amount of
redevelopment done as a fraction of the development activity assigned in the first place.
It is effected by the TD Results Discrepancy. An increase (decrease) in TD Results
discrepancy leads to an increase (decrease) in the redevelopment efforts to correct the
discrepancy. At a TD Results Discrepancy value of 0.4, the TD Redevelopment Fraction
is 0.3. It declines to a value of 0.15 when the TD Results Discrepancy decreases to a
value of 0.35. It further declines to a value of 0 as the TD Results Discrepancy decreases
to a value of 0 (NNS and NAVSEA experts). An ellipse curve (of the form x2/a2 + y2/b2
= 1) was fitted through the three points to obtain an analytical expression capturing the
effect of TD Results Discrepancy on the TD Redevelopment Fraction. An ellipse curve
was assumed based on the visual shape of the curve elicited from the clients for the effect
of TD Results Discrepancy on the TD Redevelopment Fraction. The curve fitting was
done manually. The graph depicted below (Figure 3.12) was plotted in MS Excel 2000.
70
0
0.15
0.3
0
0.1
0.2
0.3
0.4
0 0.1 0.2 0.3 0.4
TD Results Discrepancy
TD R
edev
elop
men
t Fra
ctio
n/0.16x-1*0.3-0.3y 2=
Figure 3.12 Impact of TD Results Discrepancy on TD Redevelopment Fraction
The overall equation (with inputs from NNS, NAVSEA experts) to describe the
relationship is:
TD Redevelopment Fraction = 0.3 – 0.3* 16.0/1 2epancysultsDiscrTDRe−
(3.14)
3.5.6 TD Management Effort Stock and Flow Structure
The TD Management Effort stock (man-hours) is fed into by a TD Management Initiation
Rate (man-hours/week) and is depleted by a TD Management Completion Rate (man-
hours/week). The TD Management Effort stock is an integral of the TD Management
Initiation Rate less the TD Management Completion Rate.
71
TechnologyDevelopment
Completion Rate
TD TestingCompletion Rate
TDManagement
EffortTD ManagementInitiation Rate
TD ManagementCompletion Rate
++
Funding Stability
-
TD Cost OverrunFraction
+
TD Management toDevelopment Fraction
TD Management toTesting Fraction
TD ManagementResidence Time
Figure 3.13 TD Management Effort Stock and Flow Structure
TD Management Effort (t) = TD Management Effort (0)
+ ∫ [TD Management Initiation Rate
- TD Management Completion Rate] dt
(3.15)
TD Management Effort (0) = 0
The TD Management Initiation rate is primarily driven by the Technology Development
Completion Rate and the TD Testing Completion Rate. The amount of management
done (in man-hours/week) is an additive sum of the amount of management required for
the technology development activity and the amount of management required for the
testing activity. The TD Cost Overrun Fraction is a dimensionless variable that captures
the amount of cost overruns over the total allocated funding, as a fraction of the total
allocated funding. As cost overruns increase (decrease), the management effort
72
associated with juggling the available funds among various activities occurring in the
phase increases (decreases). Or in other words, the financial management effort increases
(decreases) as the cost overruns increase (decrease). The TD Management Initiation Rate
increases at ten percent of the rate of increase in cost overruns (NNS and NAVSEA
experts). An increase (decrease) in Funding Stability leads to a decrease (increase) in the
TD Management Initiation Rate. Funding Stability (on a scale of 1 to 10; with 1 being
most unstable, increasing in stability up to 10 being the most stable) effects the value of
TD Management Initiation Rate by a multiplicative factor. At Funding Stability equals
10, the multiplicative factor is 1. At Funding Stability equals 4, the multiplicative factor
is 1.1. At Funding Stability equals 1, the multiplicative factor is 1.3. A quadratic curve
(of the form y = ax2+bx+c) was fitted through the three points to obtain an analytical
expression capturing the effect of Funding Stability on the TD Management Initiation
Rate. This equation is exactly the same as the equation for the effect of Funding Stability
on the TD Funding Inflow Rate.
1
1.1
1.3 y = 0.0056x2 - 0.0944x + 1.3889
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1 2 3 4 5 6 7 8 9 10
Funding Stability
Mul
tiplic
ativ
e Fa
ctor
Figure 3.14 Impact of Funding Stability on Management Initiation Rate
73
The overall equation (with inputs from NNS, NAVSEA experts) to describe the
relationship is:
TD Management Initiation Rate = (Technology Development Completion Rate
*TD Management to Development Fraction
+TD Testing Completion Rate
*TD Management to Testing Fraction)
* (1+0.1*TD Cost Overrun Fraction)
* (0.0056*Funding Stability2 - 0.0944
*Funding Stability + 1.3889) (3.16)
The TD Management Completion Rate is a delay function of the TD Management Effort
stock.
TD Management Completion Rate = TD Management Effort /
TD Management Residence Time (3.17)
74
3.5.7 TD Actual Costs Stock and Flow Structure
TechnologyDevelopment
Completion Rate
TD TestingCompletion Rate
TDActualCostsTD Cost
Realization Rate
+
TD ManagementCompletion Rate
+
TD Training Rate
Manhour CostRate
+
+
Figure 3.15 TD Actual Costs Stock and Flow Structure
The TD Actual Costs stock (dollars) is fed by a TD Cost Realization Rate (dollars/week).
The TD Actual Costs stock variable is an integral of the TD Cost Realization Rate.
TD Actual Costs (t) = TD Actual Costs (0)
+ ∫ [TD Cost Realization Rate] dt (3.18)
TD Actual Costs (0) = 0
The TD Cost Realization Rate is driven by the man-hour rates of technology
development, testing, and management activities. The Manhour Cost Rate (dollars/man-
hour) is the cost of labor and is used to convert the various activities’ rate units (man-
75
hours/week) into the TD Cost Realization Rate units (dollars/week). The training rate
(dollars/week) also adds to the costs realized per week.
The overall equation to describe the relationship is:
TD Cost Realization Rate = (Technology Development Completion Rate
+TD Management Completion Rate
+TD Testing Completion Rate)*Manhour Cost Rate
+ TD Training Rate (3.19)
3.5.8 TD Cost Overrun Fraction Structure
TDActualCosts
TD FundingInflow Rate
AccumulatedTD Funding
Accumulated TDFunding Inflow Rate
TD Cost OverrunFraction
-+
+
Figure 3.16 TD Cost Overrun Fraction Structure
The Accumulated TD Funding stock (dollars) is fed by the Accumulated TD Funding
Inflow Rate (dollars/week). The Accumulated TD Funding stock variable is an integral
of the Accumulated TD Funding Inflow Rate. It keeps track of the total amount of
funding received at any instant of time.
76
Accumulated TD Funding (t) = Accumulated TD Funding (0)
+ ∫ [Accumulated TD Funding Inflow Rate] dt
(3.20)
Accumulated TD Funding (0) = 0
The Accumulated TD Funding Inflow Rate is identical to the TD Funding Inflow Rate.
Accumulated TD Funding Inflow Rate = TD Funding Inflow Rate (3.21)
The TD Cost Overrun Fraction is a dimensionless variable that captures the amount of
cost overruns over the total allocated funding, as a fraction of the total allocated funding.
It has a value of 0 if the TD Actual Costs are less than the Accumulated TD Funding at a
certain instant of time. When TD Actual Costs are greater than the Accumulated TD
Funding, a cost overrun results. This cost overrun is captured in the TD Cost Overrun
Fraction variable as a fraction of the Accumulated TD Funding.
TD Cost Overrun Fraction = MAX [(TD Actual Costs-Accumulated TD Funding)/
Accumulated TD Funding, 0] (3.22)
where, MAX refers to the maximum function.
77
3.5.9 TD Training Imparted Stock and Flow Structure
TD FundingAllocation Rate
TD TrainingImparted
TD Training Rate
TD PercentageTraining
+ +
Figure 3.17 TD Training Imparted Stock and Flow Structure
The TD Training Imparted stock (dollars) is fed by a TD Training Rate (dollars/week).
The TD Training Imparted stock variable is an integral of the TD Training Rate. It keeps
track of the total amount of training imparted to the technology development technical
work force at any instant of time.
TD Training Imparted (t) = TD Training Imparted (0) + ∫ [TD Training Rate] dt
(3.23)
TD Training Imparted (0) = 0
The TD Training Rate is determined as a percentage of the TD Funding Allocation Rate.
TD Training Rate = TD Funding Allocation Rate * TD Percentage Training
(3.24)
78
3.5.10 TD Risk Structure
Complexity of NewTechnology
TD Risk
TechnologyMaturity
-
+
TD PercentageTraining
-
Initial TD Risk
+
Figure 3.18 TD Risk Structure
The TD Risk is a dimensionless variable that measures the risk associated in the
development of new technology. It is measured on a scale of 1 to 10. The TD Risk is
driven by a user-input Initial TD Risk value (input values can be 2=very low risk,
3.5=low risk, 5=average risk, 6.5=high risk, and 8=very high risk). An increase
(decrease) in the percentage training received by the work force leads to a decrease
(increase) in the TD Risk. The rate of change of TD Risk with the change in TD
Percentage Training is linear, and has different rates of change for different values of
Initial TD Risk (NNS and NAVSEA experts). The following graphical relationship
between TD Risk and TD Percentage Training at different values of Initial TD Risk
(ITDR) was elicited from NNS and NAVSEA experts.
79
21
3.5
2.25
5
3.5
6.5
4.75
8
6
0123456789
0 1 2 3 4 5
TD Percentage Training
TD R
isk
8=ITDR
5.6=ITDR
5=ITDR
5.3=ITDR
2=ITDR
Figure 3.19 Impact of TD Percentage Training on TD Risk
The equations for the above shown lines are y=8-40x (at ITDR=8), y=6.5-35x (at
ITDR=6.5), y=5-30x (at ITDR=5), y=3.5-25x (at ITDR=3.5), and y=2-20x (at ITDR=2).
As there is a symmetric change of slope across the five ITDR values, the above five
equations can be combined into one as:
y = ITDR - (20+5*(ITDR-2)/1.5) * x
where x is the TD percentage training and y is the TD risk.
An increase (decrease) in Complexity of New Technology leads to an increase (decrease)
in the TD Risk. Complexity of New Technology (on a scale of 1 to 5; with 1=very low
complexity, 2=low complexity, 3=medium complexity, 4=high complexity, 5=very high
complexity) effects the TD Risk by a multiplicative factor. When the complexity of New
Technology equals 1, the multiplicative factor is 0.8. It is assumed to follow a linear
relationship with the Complexity of New Technology up to a value of 5, where the
80
multiplicative factor is 1.2. A straight line (of the form y = mx+c) was fitted through the
two points to obtain an analytical expression capturing the effect of Complexity of New
Technology on the TD Risk. The curve fitting was done manually. The graph depicted
below (Figure 3.20) was plotted in MS Excel 2000.
0.8
1.2y = 0.1x + 0.7
0.7
0.8
0.9
1
1.1
1.2
1.3
1 2 3 4 5
Complexity of New Technology
Mul
tiplic
ativ
e Fa
ctor
Figure 3.20 Impact of Complexity of New Technology on TD Risk
An increase (decrease) in Technology Maturity leads to an decrease (increase) in the TD
Risk. Technology Maturity (on a scale of 1 to 5; with 1=very immature, 2=immature,
3=medium mature, 4=mature, and 5=very mature) effects the TD Risk by a multiplicative
factor. At Technology Maturity equals 1, the multiplicative factor is 1.2. It follows a
linear relationship with the Technology Maturity up to a value of 5, where the
multiplicative factor is 0.8. A straight line (of the form y = mx+c) was fitted through the
two points to obtain an analytical expression capturing the effect of Technology Maturity
81
on the TD Risk. The curve fitting was done manually. The graph depicted below (Figure
3.21) was plotted in MS Excel 2000.
1.2
0.8
y = -0.1x + 1.3
0.7
0.8
0.9
1
1.1
1.2
1.3
1 2 3 4 5
Technology Maturity
Mul
tiplic
ativ
e Fa
ctor
Figure 3.21 Impact of Technology Maturity on TD Risk
The overall equation (with inputs from NNS, NAVSEA experts) to describe the
relationship is:
TD Risk = (Initial TD Risk - (20+5*(Initial TD Risk-2)/1.5)*TD Percentage Training)
* (0.7+0.1*Complexity of New Technology)
* (1.3-0.1*Technology Maturity) (3.25)
82
Chapter 4. Results, Testing, Sensitivity Analysis, Validation and
Verification
The model presented in Chapter 3 was programmed in VENSIM Professional 4.0
software. The model was simulated for a specific technology whereby NNS and
NAVSEA experts provided the user-input parameters. For a different technology, there
would have been a different set of parameters. The results obtained from running the
simulation are discussed in this chapter. The dynamic hypotheses discussed in Chapter 3,
Section 3.2 were tested using the model developed in VENSIM Professional 4.0 by
varying parameters and observing the changes in the subsequent results from the
simulation. Some sensitivity analysis was also performed on the model. The simulation
was run using three sets of key parameter combinations, namely, (1) very high
technology complexity-very immature technology-no training, (2) medium technology
complexity-medium mature technology-average training, and (3) very low technology
complexity-very mature technology-high training. These were chosen to represent two
extreme condition scenarios and an average condition scenario. The results obtained are
presented in this chapter.
Validation and verification of the model seeks to enable experts and modelers to have a
better understanding of the system and share their views. This step involves the testing of
the model as to whether it replicates the behavior of the real-world system. Another
important task of validation and verification is to verify whether the variables and
parameters of the model have a meaningful concept in the real world. When models are
83
subjected to extreme conditions, their robustness is determined. A model should not just
be a means to mirror field data exactly. In addition, its elements should have a sound
conceptual basis for being in the model. The process of judging the validity of a system
dynamics model includes a number of objective tests (Richardson and Pugh, 1981, and
Sterman, 2000) that are discussed extensively in this chapter in Section 4.4.
4.1 Results
The simulation for the model was run with the following parameters:
4.1.1 Simulation Control Parameters
FINAL TIME (The final time for the simulation) = 104 Weeks
The implementation of the model was for a specific technology whereby a time horizon
of two years was used based on inputs from NNS and NAVSEA experts. For another
technology, a different time horizon would have been used.
INITIAL TIME (The initial time for the simulation) = 0 Week
TIME STEP (The time step for the simulation) = 0.03125 Week
A time step of 0.03125 week was used so as to give smooth time profiles for the different
variables in the model.
4.1.2 User defined parameters
TD Management to Development Fraction (This represents the average effort (in man-
hours) spent in management activities as a fraction of the effort (in man-hours) spent in
technology development activities.) = 0.15
84
TD Testing to Development Fraction (This represents the average effort (in man-hours)
spent in testing activities as a fraction of the effort (in man-hours) spent in technology
development activities.) = 1/6
TD Management Residence Time (This represents the average delay for management
activities.) = 1 Week
TD Management to Testing Fraction (This represents the average effort (in man-hours)
spent in management activities as a fraction of the effort (in man-hours) spent in testing
activities.) = 0.1
Initial TD Risk = 6.5 (High Risk)
The implementation of the model was for a specific technology whereby the Initial TD
Risk was set as High Risk (6.5) based on inputs from NNS and NAVSEA experts. For
another technology, a different Initial TD Risk would have been used.
TD Percentage Training (This represents the percentage of total available funds spent for
training activities) = 0.5 % of Total Funding
TD Funding Residence Time (This represents the average delay of funds in the
subsystem.) = 1 Week
85
Technology Development Effort Residence Time (This represents the average technology
development time.) = 8 Weeks
TD Testing Residence Time (This represents the average testing time.) = 4 Weeks
Complexity of New Technology = 4 (High Complexity)
The implementation of the model was for a specific technology whereby the complexity
of the technology was set as High Complexity (4) based on inputs from NNS and
NAVSEA experts. For another technology, a different Complexity of New Technology
would have been used.
Funding Stability = 8 (High funding stability)
The implementation of the model was for a specific technology whereby the Funding
Stability was set as high (8) based on inputs from NNS and NAVSEA experts. For
another technology, a different Funding Stability would have been used.
Manhour Cost Rate = 35 dollars/man-hour
NNS and NAVSEA experts provided the information that the average cost of labor was
35 dollars/man-hour.
Technology Maturity = 3 (Medium maturity)
86
The implementation of the model was for a specific technology whereby the Technology
Maturity was set as medium (3) based on inputs from NNS and NAVSEA experts. For
another technology, a different Technology Maturity would have been used.
The simulation runs showed two main modes of dynamic behavior. One dynamic
behavior was the damped oscillation observed for the variables Technology Development
Effort, TD Testing Effort, TD Management Effort, and TD Actual Costs Realization rate.
The feedback structure causing this type of dynamic behavior is identified below in
Figure 4.1.
TechnologyDevelopmentManagement
TechnologyDevelopment
Testing Effort
+
+
Cost Overrun
Actual Costs+
-
+
++
+
COM-R
DCO-B
TCO-B
MCODT-B
MCOD-B
Delay
Figure 4.1 Feedback Structure Causing Damped Oscillation
87
Oscillation arises due to negative feedback with significant time delays (Sterman, 2000).
Corrective action taken to restore the equilibrium state or to achieve the goal of the
system continues to be taken even after the equilibrium has been reached due to time
delays in identifying the effects of the actions on the system. Thus the goal is overshot.
Corrective action taken again (negative feedback loop) leads to undershooting and hence
oscillation. In a damped oscillation, as the name suggests, the oscillations die out as time
passes. Sterman (2000) says that many real world oscillatory system structures are
damped. Damped oscillatory structures are characterized by a set of negative feedback
loops as can be seen in Figure 4.1 above. As technology development effort goes up
(down), the man-hours required to carry out the actual effort goes up (down). The man-
hour effort for associated testing and management activities goes up (down) too. This
drives up (down) the actual costs of the technology development phase. As the costs
increase (decrease), the cost overruns increase (decrease). An increase (decrease) in the
cost overruns invokes corrective action taken to restore budgetary equilibrium by
effecting a decrease (increase) in the technology development activity (as a control
feedback mechanism). Corrective action taken to restore the budgetary equilibrium
continues to be taken even after the equilibrium has been reached due to time delays in
identifying the effects of the actions on the system. Thus the goal is overshot. Corrective
action taken again leads to undershooting and hence oscillation.
88
4.1.3 Technology Development Effort
CurrentTechnology Development Effort10,000
7,5005,0002,500
0Technology Development Completion Rate
2,0001,5001,000
5000
Technology Development Initiation Rate2,0001,5001,000
5000
0 26 52 78 104Time (Week)
Figure 4.2 Technology Development Stock and Flows Behavior
The amount of technology development effort to be done builds up over time. As
technology development effort that needs to be done is actually carried out, the costs
associated with the technology development activity accumulate. Once the costs
overshoot the incoming funding, the associated cost overruns dictate a slowdown in the
technology development activity to help reduce costs and thus remain within the funding
Manhours /week
Manhours /week
Manhours
89
constraints. This slowdown is observed around week 16 and confines to week 32. The
equilibrium value of the rate at which technology development needs to be done appears
to be 652 man-hours/week on running the simulation. Figure 4.2 shows the damped
oscillatory behavior observed on running the simulation. A sharp vertical jump is
observed in the Technology Development Initiation Rate at week 12. This is attributed to
the influx of the first Redevelopment activity based on the technology development and
testing having been done till then.
4.1.4 TD Testing Effort
CurrentTD Testing Effort
800600400200
0TD Testing Completion Rate
200150100
500
TD Testing Initiation Rate200150100
500
0 26 52 78 104Time (Week)
Figure 4.3 TD Testing Stock and Flows Behavior
Manhours
Manhours /week
Manhours /week
90
The amount of technology development testing effort to be done builds up over time. As
technology development testing effort that needs to be done is actually carried out, the
costs associated with the technology development testing activity accumulate. Once the
costs overshoot the incoming funding, the associated cost overruns dictate a slowdown in
the technology development testing activity to help reduce costs and thus remain within
the funding constraints. This slowdown is observed around week 20 and goes up to week
38. The equilibrium value of the rate at which TD testing needs to be done appears to be
108 man-hours/week on running the simulation. The figure (Figure 4.3) above shows the
damped oscillatory behavior observed on running the simulation.
4.1.5 TD Management Effort
CurrentTD Management Effort
200150
100
50
0TD Management Completion Rate
200150
100
50
0TD Management Initiation Rate
200150
100
50
00 26 52 78 104
Time (Week)
Figure 4.4 TD Management Stock and Flows Behavior
Manhours /week
Manhours /week
Manhours
91
The amount of technology development management effort to be done builds up over
time. As technology development management effort that needs to be done is actually
carried out, the costs associated with the technology development management activity
accumulate. Once the costs overshoot the incoming funding, the associated cost overruns
dictate a slowdown in the technology development and testing activities to help reduce
costs and thus remain within the funding constraints. This reduces the rate of
management activity associated with the technology development and testing activities.
This slowdown is observed around week 20 and goes up to week 40. On the other hand,
cost overruns entail an increase in management effort to juggle the various activities
taking place so as to remain within the funding constraints. However, this decrease in
management activity as a result of a slow down in the technology development and
testing activities dominates the increase in management activity associated with the
financial management of cost overruns. The equilibrium value of the rate at which TD
Management needs to be done appears to be 109 man-hours/week on running the
simulation. The figure (Figure 4.4) above shows the damped oscillatory behavior
observed on running the simulation.
4.1.6 TD Actual Costs
The rate at which costs are incurred in the technology development, testing and
management efforts builds up over time. Once the costs overshoot the incoming funding,
the associated cost overruns dictate a slowdown in the technology development and
testing efforts to help reduce costs and thus remain within the funding constraints. This
leads to a decrease in the cost realization rate. This slowdown is observed around week
92
20 and goes up to week 40. The equilibrium value of the rate at which costs are incurred
appears to be 30,100 dollars/week on running the simulation. The figure (Figure 4.5)
below shows the damped oscillatory behavior observed on running the simulation.
CurrentTD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.5 TD Actual Costs and Cost Realization Rate
Dollars
Dollars/week
93
The other dynamic behavior was the goal seeking observed for the variable Actual
Testing results. The feedback structure causing this type of dynamic behavior is
identified below.
Redevelopment
Actual TestingResults
Target TestingResults
Discrepancy
-
++
+
ReTDis-B
Figure 4.6 Feedback Structure Causing Goal Seeking
Goal seeking behavior arises from negative or self-controlling feedback (Sterman, 2000).
Negative feedback loops tend to oppose any changes or deviations in the state of the
system; they tend to restore equilibrium and hence are goal seeking. The rate of change
diminishes as the goal is approached, such that there is a smooth attainment of the
goal/equilibrium state of the system.
94
4.1.7 TD Redevelopment, TD Results Discrepancy, and Actual Testing Results
CurrentTD Redevelopment Fraction
0.4
0.3
0.2
0.1
0TD Results Discrepancy
0.4
0.3
0.2
0.1
00 26 52 78 104
Time (Week) Figure 4.7 TD Redevelopment Fraction and Results Discrepancy
It was assumed for the model formulation that the technology development activity yields
sixty percent of the desired results when it is carried out for the first time before any
redevelopment activity has been initiated. The first testing results yield a forty percent
discrepancy, based on which a redevelopment effort equal to thirty percent of the
technology development activity is initiated. There is an observed delay associated with
Dimension-less
week-1
95
the redevelopment activity. This is attributed to the model formulation wherein it was
assumed that the first redevelopment activity starts at week 12 once some technology
development and testing activities have taken place. Once the initiated redevelopment
activity is carried out and testing done on it, there is an increase in the actual testing
results. Thus the testing results discrepancy decreases, hence decreasing the rate at which
further redevelopment activity and testing results enhancement occurs. The figures above
(Figure 4.7) and below (Figure 4.8) show the goal seeking behavior observed on running
the simulation.
CurrentTD Actual Testing Results
1
0.9
0.8
0.7
0.6TD Results Enhancement Rate
0.04
0.03
0.02
0.01
00 26 52 78 104
Time (Week)
Figure 4.8 TD Results Enhancement Rate and Actual Testing Results
Dimension-less
week-1
96
4.2 Hypotheses Testing: Cost Performance Drivers
Hypothesis 1: A lack of appropriate training causes cost overruns and higher costs in the
technology development process.
The simulation was run at three levels of TD Training Percentages. The first simulation
was run at a TD training level being 0.5 % of the available funding. The second
simulation was run at a TD training level being 2.0 % of the available funding. The third
simulation was run at a TD training level being 4.5 % of the available funding. The
results are shown in Figure 4.9.
It is observed that the TD Actual costs, the TD Cost Realization rate, and the TD Cost
Overrun fraction decrease as the amount of training imparted (as a percentage of
available funding) is increased. The results demonstrate that the hypothesis that
increased training reduces the total costs and the cost overruns incurred is shown for the
current structure of the model.
97
training=four-and-half percenttraining=two percenttraining=five-tenths percentTD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.9 TD Costs at Three Training Levels
Dollars
Dollars/week
98
Graph for TD Cost Overrun Fraction0.4
0.2
00 16 32 48 64 80 96
Time (Week)
TD Cost Overrun Fraction : training=four-and-half percent DmnlTD Cost Overrun Fraction : training=two percent DmnlTD Cost Overrun Fraction : training=five-tenths percent Dmnl
Figure 4.10 TD Cost Overrun Fraction at Three Training Levels
Hypothesis 2: Redevelopment activities completed in the technology development
process add significantly to the cost overruns and the total costs incurred.
The simulation was run at three profiles of TD redevelopment fractions. The first
simulation was run at a profile of TD redevelopment fraction versus TD Results
discrepancy given by Equation 3.13:
TD Redevelopment Fraction = 0.3 – 0.3* 16.0/1 2epancysultsDiscrTDRe−
The TD Redevelopment Fraction is a dimensionless variable that captures the amount of
redevelopment done as a fraction of the development activity assigned in the first place.
Dimension-less
99
It is determined by the TD Results Discrepancy. An increase (decrease) in TD Results
discrepancy leads to an increase (decrease) in the redevelopment efforts to correct the
discrepancy.
0
0.15
0.3
0
0.1
0.2
0.3
0.4
0 0.1 0.2 0.3 0.4
TD Results Discrepancy
TD R
edev
elop
men
t Fra
ctio
n
/0.16x-1*0.3-0.3y 2=
Figure 4.11 Impact of TD Results Discrepancy on TD Redevelopment Fraction
Figure 4.11 (which is the same as Figure 3.12) depicts the relationship between the TD
Redevelopment Fraction and TD Results Discrepancy as elicited from NNS and
NAVSEA experts.
The second simulation was run at a thirty-five percent higher redevelopment versus
discrepancy profile,
TD Redevelopment Fraction = 1.35 * [0.3 – 0.3* 16.0/1 2epancysultsDiscrTDRe− )]
The third simulation was run at a seventy percent higher redevelopment versus
discrepancy profile,
100
TD Redevelopment Fraction = 1.70 * [0.3 – 0.3* 16.0/1 2epancysultsDiscrTDRe− )]
The results are shown below in Figure 4.12 and Figure 4.13.
redevelopment-seventy percent higherredevelopment-thirty five percent higherredevelopment-base caseTD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.12 TD Costs at Three Redevelopment Levels
Dollars
Dollars/week
101
Graph for TD Cost Overrun Fraction0.4
0.3
0.2
0.1
00 8 16 24 32 40 48 56 64 72 80 88 96 104
Time (Week)
TD Cost Overrun Fraction : redevelopment-seventy percent higher DmnlTD Cost Overrun Fraction : redevelopment-thirty five percent higher DmnlTD Cost Overrun Fraction : redevelopment-base case Dmnl
Figure 4.13 TD Cost Overrun Fraction at Three Redevelopment Levels
The results show that the increase in amount of redevelopment done as a fraction of the
technology development activity does not substantially increase the total incurred costs.
However, analysis of the effect of the increase in TD redevelopment fractions on the TD
Testing Results show (see Figure 4.14 below) a significant increase in testing results
when the redevelopment activity is increased. The conclusion that can be drawn from
this is that an increase in redevelopment fraction increases the rate at which results are
enhanced. Therefore, a faster decrease in the discrepancy between the target testing
results and the actual testing results leads to a decrease in the amount of redevelopment
that needs to be done further on. This could be a possible explanation for the dynamics
wherein the costs do not increase when redevelopment is increased. So, within the
Dimension-less
102
limitations and assumptions made for the model, insufficient understanding exists to
show the hypothesis true. On the other hand, the hypothesis testing brought forth an
interesting observation of the testing results getting better when the redevelopment
fraction is increased, without an increase in the total costs incurred. One of the reasons
for this could be that the rate at which results enhancement is set up in the model is too
fast. This means that when redevelopment is done, the results get better at a very fast rate
compared to the rate at which redevelopment is done. Therefore, redevelopment
activities do not contribute substantially to the increase in costs.
redevelopment-seventy percent higherredevelopment-thirty five percent higherredevelopment-base caseTD Actual Testing Results
1
0.9
0.8
0.7
0.6TD Results Enhancement Rate
0.06
0.045
0.03
0.015
00 26 52 78 104
Time (Week)
Figure 4.14 TD Results Enhancement Rate and Actual Results at Three Redevelopment
Levels
Dimension-less
week-1
103
Hypothesis 3: An increase in the complexity of the new technology causes an increase in
the total costs incurred.
The simulation was run at three levels of Complexity of New Technology. The first
simulation was run at a Complexity of New Technology being 1 (Very Low Complexity).
The second simulation was run at a Complexity of New Technology being 3 (Medium
Complexity). The third simulation was run at a Complexity of New Technology being 5
(Very High Complexity). The results are shown in Figure 4.15.
Complexity of new technology=5Complexity of new technology=3Complexity of new technology=1 TD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.15 TD Costs at Three Levels of Complexity
Dollars
Dollars/week
104
It is observed that the TD Actual costs and the TD Cost Realization rate increase as the
Complexity of New Technology increases. The results demonstrate that the hypothesis
that increased Complexity of New Technology increases the total costs incurred is shown
for the current structure of the model. The costs increase by about 11% when the
complexity of technology increase from 1 (very simple) to 5 (very complex).
Hypothesis 4: An increase in the maturity of the new technology causes a decrease in the
total costs incurred.
The simulation was run at three levels of Technology Maturity. The first simulation was
run at a Technology Maturity being 1 (Very Immature). The second simulation was run
at a Technology Maturity being 3 (Medium Mature). The third simulation was run at a
Technology Maturity being 5 (Very Mature). The results are shown below in Figure
4.16.
It is observed that the TD Actual costs and the TD Cost Realization rate decrease as the
Technology Maturity increases. The results demonstrate that the hypothesis that
increased Maturity of New Technology decreases the total costs incurred is shown for the
current structure of the model. A reduction of about 12% in costs is seen when the
technology maturity increases from 1 (least mature) to 5 (very mature).
105
Technology Maturity=5Technology Maturity=3Technology Maturity=1TD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.16 TD Costs at Three Levels of Technology Maturity
Dollars
Dollars/week
106
4.3 Sensitivity Analysis
The model was simulated with three sets of key parameter combinations, namely, (1)
very high technology complexity-very immature technology-no training, (2) medium
technology complexity-medium mature technology-average training, and (3) very low
technology complexity-very mature technology-high training. These were chosen to
represent two extreme condition scenarios and an average condition scenario. This
helped in understanding the sensitivity of the model results for different conditions of
technology maturity, technology complexity, and the amount of training imparted to
workers, technicians, and professionals involved in the technology development process.
The results are shown below in Figure 4.17 and Figure 4.18.
The parameter sets were:
Parameter Set 1:
Very High Complexity (Complexity of New Technology=5)
Very Immature technology (Technology Maturity=1)
No training (Percentage Training=0%)
Parameter Set 2:
Medium Complexity (Complexity of New Technology=3)
Average Mature technology (Technology Maturity=3)
Average training (Percentage Training=2.5%)
Parameter Set 3:
Very Low Complexity (Complexity of New Technology=1)
Very Mature technology (Technology Maturity=5)
High training (Percentage Training=5%)
107
Parameter Set 3Parameter Set 2Parameter Set 1Technology Development Effort10,000
7,500
5,000
2,500
0Technology Development Completion Rate
2,000
1,500
1,000
500
0Technology Development Initiation Rate
2,000
1,500
1,000
500
00 26 52 78 104
Time (Week)
Figure 4.17 Technology Development at Three Sets of Parameter Inputs
Manhours
Manhours /week
Manhours /week
108
Parameter Set 3Parameter Set 2Parameter Set 1TD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.18 TD Costs at Three Sets of Parameter Inputs
The simulated results were in agreement with expected outcomes. Parameter Set 1 was
representative of the worst-case scenario where the technology was very immature, it was
very complex, and absolutely no training was imparted to the workers, technicians, and
professionals involved in the technology development phase. This set of parameters gave
Dollars
Dollars/week
109
the highest cumulative costs and the highest values for rates at which technology
development needs to be done. Parameter Set 2 was representative of the average-case
scenario where the technology was average mature, it was average complex, and an
average level of training was imparted to the workers, technicians, and professionals
involved in the technology development phase. This set of parameters gave lower
cumulative costs and the lower values for rates at which technology development needs
to be done. Parameter Set 3 was representative of the best-case scenario where the
technology was very mature, it was very complex, and a high level of training was
imparted to the workers, technicians, and professionals involved in the technology
development phase. This set of parameters gave the lowest cumulative costs and the
lowest values for rates at which technology development needs to be done.
4.4 Testing, Verification, and Validation
Sterman (2000) outlines model testing as an iterative process that starts at the beginning
of the modeling process. A wide range of tests helps the modeler understand the
robustness and limitations of the SD model. These tests involve direct inspection of
equations, simulations of the whole model, and a qualitative or quantitative (or both)
assessment of historical fit. An important task of testing is to verify whether the variables
and parameters of the model have a meaningful interpretation in the real world. When
models are subjected to extreme conditions, their robustness is determined. A model
should not just be a means to mirror historical data exactly. In addition, its elements
should have a sound conceptual basis for being in the model.
110
Models seek to represent real world systems in a simplified manner. They differ from
reality in an infinite number of small and big ways. It is thus, impossible to “validate” a
model. Many modelers and system thinkers are of the opinion that models can be
falsified, if not verified (Bell and Senge, 1980). However, there is a flip side to that view.
Any theoretical proposition can be rescued from apparent falsification. Invoking
auxiliary hypotheses that blunt any falsification agents or data that might be applied to
the original proposition, can do this. Say, for example, a theoretical proposition fails to
satisfy one particular instance among a wide number of instances for which the
proposition has been successfully tested. An auxiliary hypothesis that the theoretical
proposition is true except in cases similar to the exceptional instance can be invoked and
the proposition guarded against complete rejection. Validation is also intrinsically social.
It seeks to enable decision-makers, modelers, and other affected groups to have a better
understanding of the system and share and concur in their views.
The process of judging the validity of a system dynamics model includes a number of
objective tests (Richardson and Pugh, 1981, and Sterman, 2000). They include:
4.4.1 Face Validity
This is an exercise to qualitatively test the fit between the stock/flow/feedback structure
of the model and the essential characteristics of the real system. The NNS and NAVSEA
experts involved in the research project confirmed the system’s causal flow, stock and
flow, and feedback structures in the final version of the model’s causal flow and stock
and flow diagrams. In fact, face validity was a continuous process. An initial causal loop
111
structure diagram was developed with inputs from and participation of NNS, NAVSEA,
and Virginia Tech System Performance Laboratory (VT SPL) group members. This
causal loop diagram was discussed and refined upon in an iterative manner over the
course of three to four modeling sessions. The evolution of the causal loop diagrams is
provided in Appendix C. A similar interactive process was followed in developing the
stock and flow structure for the model.
4.4.2 Structure Assessment Tests
Structure assessment tests are carried out to check whether the model is consistent with
the knowledge of the real system relevant to the purpose. The tests focus on the level of
aggregation and the conformance of the model to the basic physical conservation laws.
NNS and NAVSEA experts confirmed the level of aggregation of the variables and
concepts in the SD model. The model is to serve the purpose of understanding the
dynamic behavior of the system engineering implementation process as it pertains to the
introduction and implementation of new technologies in ship systems. The expert team
(NNS and NAVSEA) desired the variables and concepts in the model to be at an
aggregation level at which the model could be applied for any new technology. The
system dynamics model was built with active participation and consensus from the expert
members on the variable identification, their real-world meanings, and the concepts they
represented.
Structure assessment tests also focus on the conformance of the model to the basic
physical conservation laws. Common violations of physical law involve stocks that can
become negative. Stocks of real quantities in the model such as the amount of
112
technology development to be done (Technology Development Effort stock), the amount
of testing to be done (TD Testing Effort stock), and the amount of management required
(TD Management stock) cannot be negative. Therefore, the outflow rates from these
stocks, viz., Technology Development Completion Rate, TD Testing Completion Rate,
and TD Management Completion Rate must approach zero as the stock approaches zero.
This was tested by direct inspection of the equations (Sterman, 2000) and found to be
true.
4.4.3 Dimensional Consistency Tests
Dimensional consistency tests seek to verify that each model equation is dimensionally
consistent without the use of parameters that do not have a real world meaning. The tests
can be carried out by the direct inspection of the equations and the actual simulation of
the model (Sterman, 2000).
The model was simulated (using VENSIM software) and the simulation (as it is in the
final form) did not generate any dimensional consistency errors. Results were obtained
and they have been presented in the earlier section of this chapter. Furthermore, the
equations were directly inspected and they were found to be dimensionally consistent
without the use of any arbitrary parameters that have no real world meaning.
4.4.4 Integration Error Tests
Integration error tests are conducted to check for the model’s sensitivity to the choice of
the simulation time step. The simulation should be run with different integration time
113
steps. Ideally, the results of the model should not be sensitive to the choice of the time
step.
The simulation was run for three different integral time steps. The first simulation was
run at an Integration Time Step of 0.125 week. The second simulation was run at an
Integration Time Step of 0.03125 week. The third simulation was run at an Integration
Time Step of 0.0078125 week. The results obtained for TD Actual costs and the TD Cost
Realization Rate are shown below in Figure 4.19, Figure 4.20, and Figure 4.21.
with integration time step=0.125 weekTD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.19 TD Costs for a Simulation Run with the Integration Time Step=0.125 week
Dollars
Dollars/week
114
with integration time step=0.03125 weekTD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week) Figure 4.20 TD Costs for a Simulation Run with the Integration Time Step=0.03125
week
with integration time step=0.0078125 weekTD Actual Costs
4 M
3 M
2 M
1 M
0TD Cost Realization Rate60,000
45,000
30,000
15,000
00 26 52 78 104
Time (Week)
Figure 4.21 TD Costs for a Simulation Run with the Integration Time Step=0.0078125
week
Dollars
Dollars
Dollars/week
Dollars/week
115
The results show that the model is not sensitive to the choice of the Integration Time Step
for the simulation runs.
4.4.5 Behavior Reproduction Tests
Behavior reproduction tests are conducted to check whether the model reproduces the
behavior of interest in the system, either qualitatively, or quantitatively, or both.
The main performance metrics in the technology development process cited by NNS and
NAVSEA experts were costs and technical performance or testing results.
Graph for TD Cost Realization Rate60,000
30,000
00 16 32 48 64 80 96
Time (Week)
TD Cost Realization Rate : Current dollars/Week
Figure 4.22 TD Cost Realization Rate
Dollars/week
116
Graph for TD Actual Testing Results1
0.8
0.60 16 32 48 64 80 96
Time (Week)
TD Actual Testing Results : Current Dmnl
Figure 4.23 TD Actual Testing Results
The above two figures (Figure 4.22 and Figure 4.23) show the results obtained from
simulation runs on the model. The profiles of costs and technical performance versus
time observed in the figures above are similar in form to the behavior profiles of the
variables cost and technical performance versus time in real world as elicited from the
NNS and NAVSEA experts (see Reference Modes in Chapter 3).
Dimension-less
117
Chapter 5. Conclusion
5.1 Overview of the Results
The System Dynamics Technology Development model was simulated for a specific
technology whereby NNS and NAVSEA experts provided the user-input parameters. For
a different technology, there would have been a different set of parameters. The results
obtained on running the simulation were discussed in detail in the last chapter. The
simulation was run for two years (104 weeks), the time horizon of the technology
development process. The simulation runs showed two main modes of dynamic
behavior. One dynamic behavior was the damped oscillation observed for the variables
Technology Development Effort, TD Testing Effort, TD Management Effort, and TD
Actual Costs Realization rate. This was attributed to the presence of oscillatory
structures (in the overall causal loop and stock and flow structures) that are characterized
by a set of negative feedback loops as was seen in Figure 4.1. The other dynamic
behavior observed was the goal seeking observed for the variable Actual Testing results.
The feedback structure causing this type of dynamic behavior was identified in Figure
4.6.
5.2 Verification of the Dynamic Hypotheses
The dynamic hypotheses discussed in Chapter 3, Section 3.2 were tested using the SD
model developed in VENSIM Professional 4.0 by varying parameters and observing the
changes in the subsequent results from the simulation. The first hypothesis was that a
lack of appropriate training causes cost overruns and higher costs in the technology
development process. From the hypothesis testing simulation results, it was observed
118
that the TD Actual costs, the TD Cost Realization rate, and the TD Cost Overrun fraction
decreased as the amount of training imparted (as a percentage of available funding) was
increased. The results demonstrated that the hypothesis that increased training reduces
the total costs and the cost overruns incurred was shown for the current structure of the
model.
The second hypothesis was that redevelopment activities completed in the technology
development process add significantly to the cost overruns and the total costs incurred.
From the hypothesis testing simulation results, it was observed that an increase in the
amount of redevelopment done as a fraction of the technology development activity did
not substantially increase the total incurred costs. So, within the limitations and
assumptions made for the current structure of the model, insufficient understanding exists
to prove the hypothesis true. However, analysis of the effect of the increase in TD
redevelopment fractions on the TD Testing Results showed a significant increase in
testing results when the redevelopment activity was increased.
The third hypothesis was that an increase in the complexity of the new technology causes
an increase in the total costs incurred. From the hypothesis testing simulation results, it
was observed that the TD Actual costs and the TD Cost Realization rate increased as the
Complexity of New Technology increased. The results demonstrated that the hypothesis
that increased Complexity of New Technology increases the total costs incurred was
shown for the current structure of the model.
119
The fourth hypothesis was that an increase in the maturity of the new technology causes a
decrease in the total costs incurred. From the hypothesis testing simulation results, it was
observed that the TD Actual costs and the TD Cost Realization rate decreased as the
Technology Maturity increased. The results demonstrated that the hypothesis that
increased Maturity of New Technology decreases the total costs incurred was shown for
the current structure of the model.
5.3 Policy Suggestions
Based on the results and testing done on the SD model, certain policy suggestions can be
made. Training was observed to have a substantial impact on cost reductions. It would
make sense as a policy shift to devote more resources and attention to training in large
technology development projects. This issue gains much more importance in the context
of new technologies. Technical issues continuously challenge new technology
development processes during the entire time horizon of the process due to the lack of a
historical perspective and understanding with respect to the new technology. As seen by
the hypothesis testing on the impact of training on costs and cost overruns, it would be a
wise policy shift to allocate more funds and effort to training activities and is usually
done in current technology development projects.
The thrust of new technology implementations should be towards spending more funds
on academic research to try and understand the technology better before attempting to
implement it on aircraft carriers. It was seen from the hypothesis testing on impact of
complexity of technology on costs and cost overruns that a simpler technology incurs
120
lower costs than a complex one. It can be conjectured that pursuing more academic
research on a new technology would enhance its understanding and thus reduce the
complexity of the technology, which in turn would reduce the costs incurred during its
implementation on aircraft carriers.
Another policy shift could be in deciding when to actually implement the technology on
the aircraft carriers. New technology implementations that are mainly in
telecommunications and Information Technology sectors are often time sensitive. They
need to be incorporated fast to keep pace with the technological challenges and needs.
However, the drawback to this is that an immature technology entails increased costs as
was seen in Chapter 4 during the hypothesis testing of impact of technology maturity on
costs. The other critical issue is that some technologies become obsolete very fast.
Therefore, if a technology implementation were not very sensitive to when exactly it was
done, it would be wise to postpone the process. It would serve as a double-edged
advantage. The first is that increased technology maturity would reduce costs incurred.
The second is that if the technology still needs to be implemented after a substantially
postponed schedule, it would indicate that the technology is relatively not very volatile in
terms of how soon it becomes obsolete.
5.4 Future Issues
System Dynamics modeling is inherently iterative. One of the main issues in SD
modeling is the level of abstraction in the model (Sterman 2000). The assumptions made
in a system dynamics model determine what concepts have been included in the model
121
and what concepts and variables that have been left out. Furthermore they also determine
the level of detail to which the concepts are treated in the model. One of the main issues
for further research is going to a more detailed level of abstraction for the Technology
Development variable. In its current form in the model formulation, technology
development is at a very high level in the scheme of things, and it includes activities like
up-front research and development, adaptability studies, development of drawing
specifications, development of component drawings, and prototype development all
grouped together. Breaking down the Technology Development variable into the
daughter activities and studying their dynamics individually could be one of the future
research issues.
Another issue for future research would be incorporating schedule overruns in the model.
The current model structure ignores the issue of schedule overruns. Similar to cost
overruns, schedule overruns could form part of a significant feedback mechanism within
the model affecting other key variables. This is a definite research area to be explored.
The current structure of the model formulation assumes a steady inflow of funding per
week for activities taking place in the Technology Development phase. In real-life, the
funding profile could be conjectured to follow a more erratic time profile than a steady
constant flow. It would be interesting to incorporate a new funding inflow profile into
the model, and observe and analyze the subsequent dynamics of the behavior of the
system based on the altered model formulation.
122
Another future research issue deals with the “effort” variables in the current model
formulation. The “effort” variables in the model as of now are assumed to be at a very
high level in the system and are translated as the number of man-hours expended in the
respective activities and efforts. However, at a further level of detail, one could think of
these efforts to be comprised mainly of labor and material flows. Thus separating the
efforts into labor and material streams and studying them as co-flows will be another
issue and area for further research.
123
References
Abdel-Hamid, T. K., and S. E. Madnick (1991). Software Project Dynamics: An
Integrated Approach, Prentice Hall: Englewood Cliffs, NJ.
Andersen, D. F., G. P. Richardson, and J. A. M. Vennix (1997). Group model Building:
adding more science to the craft. System Dynamics Review, Vol. 13, no. 2, pp. 187-201.
Bell, J., and P. Senge (1980). Elements of the System Dynamics Method, Pegasus
Communications: Waltham, MA.
Cooper, K. G. (1980). Naval Ship Production: A Claim Settled and a Framework Built.
Interfaces, Vol. 10, no. 6, pp. 30-36.
Drew, D. R. (1995). System Dynamics Modeling and Application, Course notes, ENGR
5104, Virginia Tech.
Ford, D., and J. D. Sterman (1998). Expert knowledge elicitation for improving mental
and formal models. System Dynamics Review, Vol. 14, no. 4, pp. 309-340.
Forrester, J. W. (1958). Industrial Dynamics: A Major Breakthrough for Decision
Makers. Harvard Business Review, July-August, pp. 37-66.
124
Forrester, J. W. (1968). Principles of Systems, Wright-Allen Press, Inc.: Cambridge,
MA.
Goodman, P., and T. Griffith, (1991). A process approach to the implementation of new
technology. Journal of Engineering and Technology Management, Vol. 8, pp. 261-285.
Griffith, T. L. (1999). Technology features as triggers for sensemaking. Academy of
Management Review, Vol. 24, pp. 472-488.
Griffith, T. L., and G. B. Northcraft (1993). Promises, pitfalls, and paradox: Cognitive
elements in the implementation of new technology. Journal of Managerial Issues, Vol.
5, pp. 465-482.
Griffith, T. L., R. F. Zammutoo, and L. Aiman-Smith (1999). Why new technologies
fail. Industrial Management, Vol. 41, no. 3, pp. 29-34.
Iansiti, M. (1997). Technology Integration: Making critical choices in a dynamic world,
Harvard Business School Press: Boston, MA.
Richardson, G. P. (1996). Problems for the future of system dynamics. System
Dynamics Review, Vol. 12, no. 2, pp. 141-157.
125
Richardson, G. P., and A. L. Pugh (1981). Introduction to System Dynamics Modeling
with Dynamo, MIT Press: Cambridge, MA.
Richardson, G. P. and D. F. Andersen (1995). Teamwork in Group Model Building.
System Dynamics Review, Vol. 11, no. 2, pp. 113-137.
Roberts, E. B. (1964). The Dynamics of Research and Development, Harper & Row:
New York.
Roberts, N., D. F. Andersen, R. M. Deal, M. S. Grant, and W. A. Shaffer (1983).
Introduction to Computer Simulation: The System Dynamics Modeling Approach,
Addison-Wesley: Reading, MA.
Sastry, M. A., and Sterman, J. D. (1992). Desert Island Dynamics: An Annotated Survey
of the Essential System Dynamics Literature. Proceedings of the 1993 Internatioanl
System Dynamics Conference, Cancun, Mexico: pp. 466-475, System Dynamics Society:
Lincoln, MA.
Sink, D. S. (1983). Using the Nominal Group Technique Effectively. National
Productivity Review, Spring, pp. 173-183.
Sterman, J. D. (1994). Learning in and about complex systems. System Dynamics
Review, Vol. 10, 2-3, pp. 291-330.
126
Sterman, J. D. (2000). Business Dynamics: Systems Thinking and Modeling for a
Complex World, Irwin McGraw-Hill: Boston, MA.
Vennix, J. A. M., D. F. Andersen, G. P. Richardson, and J. Rohrbaugh (1992). Model-
Building for group decision support: Issues and alternatives in knowledge elicitation.
European Journal of Operational Research, Vol. 59, pp 28-41.
127
Appendix A. Glossary
Condition Readiness: The Navy has three Conditions of Readiness. A Condition I battle
readiness requires all operational systems fully manned and operating and no
maintenance actions except urgent repairs. A Condition II limited action requires all
possible operating stations manned and ready and urgent preventive maintenance actions
and administrative functions performed. A Condition III cruising requires operating
stations manned only as necessary and normal underway-preventive maintenance actions
and administration functions performed.
Endogenous: It is defined as “arising from within”, that is, from inside the boundary of
the model.
Exogenous: It is defined as “arising from without”, that is, from outside the boundary of
the model.
Facilitator: A facilitator conducts and manages the group session, with an aim of
facilitating the expression of views of all the key players involved in the modeling
exercise.
First Order Delay: A first order delay is a stock and flow dynamics characterized by the
outflow rate being proportional to the stock size. Some units have lower residence time
128
than the average delay time while others have greater residence times than the average
delay time.
Gatekeeper: A gatekeeper is a contact person within the target organization who helps
select the appropriate people to work within the organization with, helps plan the
modeling sessions, and works closely with the modeling group.
Mapping System Structure: It is the process of mapping the boundary of the model and
representing its causal structure.
Modeler: A modeler’s role is to be a keen observer in the modeling exercise, listening
and capturing the ideas and views of different session members and providing inputs on
the technical modeling aspects as and when required.
Model Boundary Chart: It lists the key endogenous and exogenous variables of the
model.
Process Coach: A process coach ensures that the modeling group does not stray
substantially from the modeling agenda, and provides feedback to the facilitator in terms
of the effectiveness of the modeling process being followed.
Recorder: A recorder takes detailed notes on the various session activities and
developments.
129
Refresh: A technology refresh refers to an upgrade of the technology features or
operational capabilities.
130
Appendix B. System-subsystem structure
System
The system is the system engineering process that is responsible for the development,
implementation, maintenance, upgrade, and retirement of new technologies.
Subsystems
Four subsystems were identified for the overall system, namely, Program Management
(Subsystem #1), Technology Development (Subsystem #2), Technology Integration
(Subsystem #3), and Operations, Support & Disposal (Subsystem #4)
Subsystem #1: Program Management
Program Management is responsible for managing the technology implementation in
response to validated operational requirements. It provides the coordination and direction
of activities required to develop, integrate, operate, maintain, and dispose the new
technology. It oversees cost and secures funds for the technology implementation
activities.
Subsystem #2: Technology Development
In the Technology Development phase, new technologies are developed and/or assessed.
The various activities for this subsystem include R&D technology development activities
(technology research, requirements definition, specification development, engineering,
modeling and simulation, drawing development, hardware and software development,
131
system architecture development, and project management), Project Management
activities, and Testing activities.
Subsystem #3: Technology Integration
This subsystem is primarily responsible for the integration of the new technology into the
ship operations. The various activities for this subsystem include Engineering and
Design (feasibility evaluation of new designs, design, development and implementation
of design specifications, development of drawings, and managing engineering changes),
Procurement and Fabrication (Purchasing of parts, components, hardware, and
software), Assembly, Installation and Integration (Integration of new technologies with
existing technologies and operations, and technology installation), Project Management
activities, and Testing activities
Subsystem #4: Operations, Support & Disposal
This subsystem is primarily responsible for all the activities necessary for the operations,
support and disposal of the new technology. The various activities for this subsystem
include Shipboard Maintenance (operation and maintenance of the systems onboard the
ship), Organizational and Intermediate Maintenance (preventive and corrective actions
requiring calibration, repair, and replacement of parts, components, and assemblies),
Depot Maintenance (performing major overhauls or maintenance on a ship and
associated equipment at centralized repair depots, or contractor repair facilities),
Other/Indirect/Sustaining Support (course training for the ships crew to enable them to
perform assigned maintenance and operational tasks), and Operations and Support
132
Disposal (compliance with Environmental/HAZAMAT laws and regulations and with
handling and disposal of hazardous material during the operating life of the ship).
133
Overall System/Subsystem Diagram
TechnologyDevelopment
TechnologyIntegration
ProgramManagement
Operations,Support, and
Disposal
TechnologyDevelopment Funding
TechnologyDevelopment Cost
Overruns
Request for additionalfunding for Technology
Development
Request for additionalfunding for Technology
Integration
Request for more funding forOperations, Support, and
Disposal
Integration RiskOperational Risk
TechnologyModernization
Technology IntegrationCost Overruns
Operations andSupport Cost
Overruns
TechnologyIntegration Funding
Operations, Support,and Disposal Funding
134
Appendix C. Evolution of Causal Loop Diagrams
I. Causal loop diagram on April 06, 2001
High LevelRequirements
TechnologyMaturity
Specifications
TechnologyDevelopment Risk
Complexity of newTechnology
ProjectManagement
SystemDevelopment
Integration RiskEvaluation
Funding
+
+
+
Testing Effort
Redevelopment
Technology DevelopmentPerformance Assessment
Funding Stability
Actual TestingResults
+
++
++
+
+
-
+-
+
Target TestingResults
Discrepancy -
+
+
-
-
-
135
II. Causal loop diagram on April 19, 2001
High LevelRequirements
TechnologyMaturity
Specifications
TechnologyDevelopment Risk
Complexity of newTechnology
ProjectManagement
TechnologyDevelopment
Integration Risk
Funding
+
+
+
Testing Effort
Redevelopment
Technology DevelopmentPerformance Assessment
Funding Stability
Actual TestingResults
+
++
++
+
+
-
+-
+
Target TestingResults
Discrepancy -
+
+
-
-
-
+
+
136
III. Causal loop diagram on May 16, 2001
High LevelRequirements
TechnologyMaturity
TechnologyDevelopment Risk
Complexity of newTechnology
TechnologyDevelopmentManagement
TechnologyDevelopment
Integration Risk
Funding
Testing Effort
Redevelopment
Technology DevelopmentPerformance Assessment
Funding Stability
Actual TestingResults
+
++
+
-
+ -
Target TestingResults
Discrepancy
-
++
-
-
-
+
+
+
+
+
+
Training
+
-
Cost Overrun Actual Costs+
-
-
+
+ +
+
COM-R
DCO-B TC
O-B
MCODT-B
MCOD-B
DTrR-B
ReTDis-B
137
IV. Causal loop diagram on May 29, 2001
TechnologyMaturity
TechnologyDevelopment Risk
Complexity of newTechnology
TechnologyDevelopmentManagement
TechnologyDevelopment
Integration Risk
Funding
Testing Effort
Redevelopment
TechnologyDevelopmentPerformance
Funding Stability
Actual TestingResults
+
+
+
-
+
Target TestingResults
Discrepancy
-
++
-
-
+
+
+
+
+
Training
-
Cost Overrun Actual Costs+
-
-
+
+ +
+
COM-R
DCO-B TC
O-B
MCODT-B
MCOD-B
ReTDis-B
+
+
+
138
Vita
Pavinder Monga
Pavinder works as a Senior Policy Analyst at First USA Bank in Delaware, USA. He is
involved with framing credit policies for new credit card applicants, analyzing portfolio
performances, and making strategy recommendations for credit market expansion. His
interests and expertise are in the areas of strategy decision management, derivatives risk
management, data analytics and portfolio restructuring. Pavinder received his Master of
Science degree in Operations Research from the department of Industrial and Systems
Engineering at Virginia Tech., Blacksburg, U.S.A. He received his Bachelor of
Technology (Honors) degree in Chemical Engineering from the Indian Institute of
Technology, Kharagpur, India.